research method semantic network analysis

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Publications
Account settings
My Bibliography
Collections
Citation manager

Save citation to file

Email citation, add to collections.

Create a new collection
Add to an existing collection

Add to My Bibliography

Your saved search, create a file for external citation management software, your rss feed.

Search in PubMed
Search in NLM Catalog
Add to Search

Semantic network analysis (SemNA): A tutorial on preprocessing, estimating, and analyzing semantic networks

Affiliations.

1 Department of Neurology, University of Pennsylvania.
2 Faculty of Industrial Engineering and Management, Technion Israel Institute of Technology.
PMID: 34941329
DOI: 10.1037/met0000463

To date, the application of semantic network methodologies to study cognitive processes in psychological phenomena has been limited in scope. One barrier to broader application is the lack of resources for researchers unfamiliar with the approach. Another barrier, for both the unfamiliar and knowledgeable researcher, is the tedious and laborious preprocessing of semantic data. We aim to minimize these barriers by offering a comprehensive semantic network analysis pipeline (preprocessing, estimating, and analyzing networks), and an associated R tutorial that uses a suite of R packages to accommodate the pipeline. Two of these packages, SemNetDictionaries and SemNetCleaner , promote an efficient, reproducible, and transparent approach to preprocessing linguistic data. The third package, SemNeT , provides methods and measures for estimating and statistically comparing semantic networks via a point-and-click graphical user interface. Using real-world data, we present a start-to-finish pipeline from raw data to semantic network analysis results. This article aims to provide resources for researchers, both the unfamiliar and knowledgeable, that reduce some of the barriers for conducting semantic network analysis. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

PubMed Disclaimer

LinkOut - more resources

Full text sources.

American Psychological Association
Ovid Technologies, Inc.
Citation Manager

NCBI Literature Resources

MeSH PMC Bookshelf Disclaimer

The PubMed wordmark and PubMed logo are registered trademarks of the U.S. Department of Health and Human Services (HHS). Unauthorized use of these marks is strictly prohibited.

Main content

Links to this project.

Title	Authors

Loading wiki pages...

Add important information, links, or images here to describe your project.

Add new wiki page

Wiki syntax help.

The wiki uses the Markdown syntax. For more information and examples, go to our Guides.

Page permissions have changed

Your browser should refresh shortly…

Renaming wiki...

Wiki page deleted

Press Confirm to return to the project wiki home page.

Connected to the collaborative wiki

This page is currently connected to the collaborative wiki. All edits made will be visible to contributors with write permission in real time. Changes will be stored but not published until you click the "Save" button.

Connecting to the collaborative wiki

This page is currently attempting to connect to the collaborative wiki. You may continue to make edits. Changes will not be saved until you press the "Save" button.

Collaborative wiki is unavailable

The collaborative wiki is currently unavailable. You may continue to make edits. Changes will not be saved until you press the "Save" button.

Browser unsupported

Your browser does not support collaborative editing. You may continue to make edits. Changes will not be saved until you press the "Save" button.

Start managing your projects on the OSF today.

Free and easy to use, the Open Science Framework supports the entire research lifecycle: planning, execution, reporting, archiving, and discovery.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

View all journals
Explore content
About the journal
Publish with us
Sign up for alerts
Open access
Published: 20 July 2023

Topological properties and organizing principles of semantic networks

Gabriel Budel 1 na1 ,
Ying Jin 1 na1 ,
Piet Van Mieghem 1 &
Maksim Kitsak 1

Scientific Reports volume 13 , Article number: 11728 ( 2023 ) Cite this article

2142 Accesses

1 Citations

3 Altmetric

Metrics details

Complex networks
Computational science
Information technology

Interpreting natural language is an increasingly important task in computer algorithms due to the growing availability of unstructured textual data. Natural Language Processing (NLP) applications rely on semantic networks for structured knowledge representation. The fundamental properties of semantic networks must be taken into account when designing NLP algorithms, yet they remain to be structurally investigated. We study the properties of semantic networks from ConceptNet, defined by 7 semantic relations from 11 different languages. We find that semantic networks have universal basic properties: they are sparse, highly clustered, and many exhibit power-law degree distributions. Our findings show that the majority of the considered networks are scale-free. Some networks exhibit language-specific properties determined by grammatical rules, for example networks from highly inflected languages, such as e.g. Latin, German, French and Spanish, show peaks in the degree distribution that deviate from a power law. We find that depending on the semantic relation type and the language, the link formation in semantic networks is guided by different principles. In some networks the connections are similarity-based, while in others the connections are more complementarity-based. Finally, we demonstrate how knowledge of similarity and complementarity in semantic networks can improve NLP algorithms in missing link inference.

Topological methods for data modelling

Structural differences in the semantic networks of younger and older adults

Embedding the intrinsic relevance of vertices in network analysis: the case of centrality metrics

Introduction.

Due to the explosive increase in the availability of digital content, the demand for computers to efficiently handle textual data has never been greater. Large amounts of data and improved computing power have enabled a vast amount of research on Natural Language Processing (NLP). The goal of NLP is to allow computer programs to interpret and process unstructured text. In computers, text is represented as a string, while in reality, human language is much richer than just a string. People relate text to various concepts based on previously acquired knowledge. To effectively interpret the meaning of a text, a computer must have access to a considerable knowledge base related to the domain of the topic 1 .

Semantic networks can represent human knowledge in computers, as first proposed by Quillian in the 1960s 2 , 3 . ‘Semantic’ means ‘relating to meaning in language or logic’ and a semantic network is a graph representation of structured knowledge. Such networks are composed of nodes, which represent concepts ( e.g. , words or phrases), and links, which represent semantic relations between the nodes 4 , 5 . The links are tuples of the format (source, semantic relation, destination) that encode knowledge. For example, the information that a car has wheels is represented as (car, has, wheels) . Figure 1 shows a toy example of a semantic network as the subgraph with the neighborhood around the node car .

Toy example of a semantic network with six concepts and five semantic relations of four different types.

The past two decades have witnessed a rise in the importance of NLP applications 6 , 7 , 8 . For instance, Google introduced Google Knowledge Graph to enhance their search engine results 9 . A knowledge graph is a specific type of semantic network, in which the relation types are more explicit 10 , 11 . Voice assistants and digital intelligence services, such as Apple Siri 12 and IBM Watson 13 , use semantic networks as a knowledge base for retrieving information 14 , 15 . As a result, machines can process information in raw text, comprehend unstructured user input, and achieve the goal of communicating with users, all up to a certain extent. Recently, OpenAI made a great leap forward in user-computer interaction with InstructGPT, better known to the general public as ChatGPT 16 .

Language is a complex system with diverse grammatical rules. To grasp the meaning of a sentence, humans leverage their natural understanding of language and concepts in contexts. Language is still poorly understood from a computational perspective and, hence, it is difficult for computers to utilize similar strategies. Namely, machines operate under unambiguous instructions that are strictly predefined and structured by humans. Though we can argue that human languages are structured by grammar, these grammatical rules often prove to be ambiguous 17 . After all, in computer languages, there are no synonyms, namesakes, or tones that can lead to misinterpretation 18 . Computers rely on external tools to enable the processing of the structure and meaning of texts.

In this paper, we conduct systematic analyses of the topological properties of semantic networks. Our work is motivated by the following purposes:

Understand fundamental formation principles of semantic networks.

In many social networks connections between nodes are driven by similarity 19 , 20 , 21 , 22 . The more similar two nodes are in terms of common neighbors, the more likely they are connected. Thanks to the intensive study of similarity-based networks, many successful tools of data analysis and machine learning were developed, such as link prediction 23 and community detection 24 . These tools may not work well for semantic networks, because words in a sentence do not necessarily pair together because of similarity. Sometimes, two words are used in conjunction because they have complementary features. Therefore, we study the principles that drive the formation of links in semantic networks.

Document language-specific features.

Languages vary greatly between cultures and across time 25 . Two languages that originate from two different language families can differ in many types of features since they are based on different rules. It is natural to conjecture that there exist diverse structures in semantic networks for different languages.

Better inform NLP methods.

Although there have been numerous real-world NLP applications across various domains, existing NLP technologies still have limitations 26 . For example, processing texts from a language where single words or phrases can convey more than one meaning is difficult for computers 27 , 28 . Existing, successful algorithms built on top of semantic networks are usually domain-specific, and designing algorithms for broader applications remains an open problem. To design better language models that can handle challenges such as language ambiguity, we first need to gain a better understanding of the topological properties of semantic networks.

Previous studies on semantic networks focused on a few basic properties and relied on multiple datasets with mixed semantic relations, which we discuss in detail in the ‘Related work’ section. Therefore, it is difficult to compare the results within one study and between two different studies. To our knowledge, there has been no systematic and comprehensive analysis of the topological properties of semantic networks at the semantic relation level.

To sum up, the main objective of this paper is to understand the structure of semantic networks. Specifically, we first study the general topological properties of semantic networks from a single language with distinct semantic relation types. Second, we compare semantic networks with the same relation type between different languages to find language-specific patterns. In addition, we investigate the roles of similarity and complementarity in the link formation principles in semantic networks.

The main contributions of this paper include:

We study the topological properties of seven English semantic networks, each network defined by a different semantic relation ( e.g. , ‘Is-A’ and ‘Has-A’). We show that all networks possess high sparsity and many possess a power-law degree distribution. In addition, we find that most networks have a high average clustering coefficient, while some networks show the opposite.

We extend the study of the topological properties of semantic networks to ten other languages. We find non-trivial structural patterns in networks from languages that have many grammatical inflections. Due to the natural structure of grammar in these languages, words have many distinct inflected forms, which leads to peaks in the density of the degree distribution and results in deviations from a power law. We find this feature not only in inflecting languages but also in Finnish, which is classified as agglutinating.

We study the organizing principles of 50 semantic networks defined by different semantic relations in different languages. We quantify the structural similarity and complementarity of semantic networks by counting the relative densities of triangles and quadrangles in the graphs, following a recent work by Talaga and Nowak 29 . Hereby, we show to what extent these networks are similarity- or complementarity-based. We find that the connection principles in semantic networks are mostly related to the type of semantic relation, not the language origin.

This paper is organized in the following manner: In the ‘Related work’ section, we provide a brief overview of the previous work on the properties of semantic networks. In the ‘General properties of semantic networks’ section, we study the general topological properties of seven English semantic networks. In the ‘Language-specific properties’ section, we compare the properties of semantic networks between 11 different languages. The section ‘Similarity and complementarity in semantic networks’ deals with the fundamental connection principles in semantic networks. We measure and compare the structural similarity and complementarity in the networks in this study and we discuss the patterns that arise. Finally, we summarize our conclusions and findings and give recommendations for future research in the ‘Discussion’ section.

Related work

Due to the growing interest in semantic networks, related studies were carried out in a wide range of different fields. Based on our scope, we focus on two main aspects in each work: the topological properties that were analyzed in the study and the dataset that was used in the analysis (i) and the universal and language-specific patterns which were found and discussed (ii).

The majority of semantic networks literature is centered around three link types: co-occurrence, association and semantic relation. In a co-occurrence network, sets of words that co-occur in a phrase, sentence or text form a link. For association networks, participants in a cognitive-linguistic experiment are given a word and asked to give the first word that they think of. There are several association datasets, one example is the University of South Florida Free Association Norms 30 . Semantic relations are relations defined by professionals like lexicographers, typical examples are synonym, antonym, hypernym and homonymy. The specific instances of the semantic relations are also defined by the lexicographers or extracted computationally from text corpora.

In 2001, Ferrer-i-Cancho and Sole 31 studied undirected co-occurrence graphs constructed from the British National Corpus dataset 32 . They measured the average distance between two words and observed the small-world property, which was found in many real-world networks 33 . Motter et al. 34 analyzed an undirected conceptual network constructed from an English Thesaurus dictionary 35 . They focused on three properties: sparsity (small average degree), average shortest path length and clustering. That same year, Sigman and Cecchi 36 studied undirected lexical networks extracted from the noun subset of WordNet 37 , where the nodes are sets of noun synonyms. They grouped networks by three semantic relations: antonymy, hypernymy and meronymy. A detailed analysis of characteristic length (the median minimal distance between pairs of nodes), degree distributions and clustering of these networks were provided. Semantic networks were also found to possess the small-world property of sparse connectivity, short average path length, and strong local clustering 34 , 36 .

Later, Steyvers and Tenenbaum 38 performed statistical analysis of 3 kinds of semantic networks: word associations 30 , WordNet and Roget’s Thesaurus 39 . Apart from the above-mentioned network properties, they also considered network connectedness and diameter. They pointed out that the small-world property may originate from the scale-free organization of the network, which exists in a variety of real-world systems 40 , 41 , 42 .

As for patterns across different languages, Ferrer-i-Cancho et al. 43 built syntactic dependency networks from corpora (collections of sentences) for three European languages: Czech, German and Romanian. They showed that networks from different languages have many non-trivial topological properties in common, such as the small-world property, a power-law degree distribution and disassortative mixing 44 .

Existing studies have identified some general network properties in semantic networks such as the small-world property and power-law degree distributions. However, the datasets used in these studies are often different, sometimes even within the same study, rendering direct comparison of results difficult. Some used associative networks generated from experiments and some studied thesauri that were manually created by linguists. In addition, most of the research performed consists of coarse-grained statistical analyses. Specifically, different semantic relations were sometimes treated as identical and the subset of included nodes was often limited ( e.g. , only words and no phrases or only nouns). Further, there are only very few studies on semantic networks from languages other than English.

Therefore, our analyses focus on semantic networks with different semantic relations (link types) from a single dataset. We consider networks defined by a specific link type, make these networks undirected and unweighted and compare the structural properties between networks with different link types. In addition, we apply similar analyses to semantic networks with the same link type across different languages. Furthermore, we investigate the roles that similarity and complementarity play in the formation of links in semantic networks.

General properties of semantic networks

To gain an understanding of the structure of semantic networks, we first study their general topological properties. We introduce the main characteristics of the dataset that we use throughout this study, ConceptNet 45 , in the section ‘Data’ in SI. Next, we list the semantic relations that define the networks in this study in Supplementary Table S1 . The overview of the semantic networks is given in Supplementary Table S2 . In this section, we compute various topological properties of these networks related to connectedness, degree, assortative mixing and clustering. We summarize the overall descriptive statistics of the semantic networks in Supplementary Table S3 .

Connectedness

We measure the connectedness of a network by the size of the largest connected component and the size distribution of all connected components. The complete component size distributions of the English semantic networks are shown in Supplementary Figure S2 . Supplementary Table S4 lists the sizes of the largest connected components (LCCs) in absolute numbers as well as relative to the network size. The same statistics are computed after degree-preserving random rewiring of the links for comparison 44 . The purpose of random rewiring is to estimate the value of a graph metric that could be expected by chance, solely based on the node degrees (see SI for details on the rewiring process).

Based on the percentages of nodes in the LCC, all seven semantic networks are not fully connected. The networks ‘Is-A’, ‘Related-To’ and ‘Union’ are almost fully connected given that their LCCs contain over 90% of nodes. Networks ‘Has-A’, ‘Part-Of’, ‘Antonym’ and ‘Synonym’ are largely disconnected, with the percentages of nodes in their LCCs ranging from 22% to 65%. Most of the rewired networks are more connected than the corresponding original networks, especially networks ‘Antonym’ and ‘Synonym’. In other words, the majority of our semantic networks are less connected than what could be expected by chance. For networks ‘Related-To’ and ‘Union’, the percentage of nodes in the LCC remains almost unchanged, while the ‘Is-A’ network is more connected than expected.

Degree distribution

Figure 2 shows that the densities \(\Pr [D=k]\) of the degree distributions of our seven English semantic networks all appear to approximately follow power laws in the tail visually. A more rigorous framework for assessing power laws was proposed by Voitalov et al. 46 , who consider networks to have a power-law degree distribution if \(\Pr [D=k] = \ell (k) k^{-\gamma }\) for a slowly varying function \(\ell (k)\) , see the section ‘Consistent power-law exponent estimators’ in SI. Figure 2 includes the estimates \({\hat{\gamma }}\) based on the slopes of the densities \(\Pr [D=k]\) on a log-log scale, along with the three consistent estimators from the framework of Voitalov et al. 46 , 47 . According to these estimators, the degree sequences of 5 out of the 7 networks are power-law. The degree sequences of the ‘Synonym’ and ‘Antonym’ networks are hardly power-law because at least one of the \({\hat{\gamma }} > 5\) and therefore the estimated exponents are not listed.

For most networks, the estimated exponent \({\hat{\gamma }}\) lies between 2 and 3. Therefore, most semantic networks are scale-free, except for the ‘Synonym’ and ‘Antonym’ networks. In the literature, semantic networks were also found to be highly heterogeneous 38 , 48 . Moreover, the word frequencies in several modern languages were found to follow power laws 49 . In the section ‘Language-specific properties’, we will see that the ‘Synonym’ and ‘Antonym’ networks in most considered languages are hardly power-law or not power-law networks.

The heterogeneity in the degree distribution seems natural for networks such as the ‘Is-A’ network: there are many specific or unique words with a small degree that connect to only a few other words, while there are also a few general words that connect to almost anything, resulting in a large degree. Examples of general words with a large degree are ‘plant’ and ‘person’, while specific words like ‘neotectonic’ and ‘cofinance’ have a small degree. Our results show that many semantic networks have power-law degree distributions, like many other types of real-world networks 50 , 51 , 52 .

Degree distribution densities \(\Pr [D=k]\) of the LCCs of the seven English semantic networks. The data is scaled by powers of 1000 to better visualize the power law in each density. The corresponding estimated power-law exponents \({\hat{\gamma }}\) are shown if there is a power law, \(\Pr [D=k] \approx \ell (k) k^{-\gamma }\) . The degree sequences of the networks ‘Antonym’ and ‘Synonym’ were estimated to be hardly power-law because at least one of the \({\hat{\gamma }} > 5\) . The data are logarithmically binned to suppress noise in the tails of the distributions, see the section ‘Power-law degree distributions’ in SI for details on how the power-law densities are processed and the power-law exponent estimation procedures.

Degree assortativity

A number of measures have been established to quantify degree assortativity, such as the degree correlation coefficient \(\rho _D\) and the Average Nearest Neighbor Degree (ANND) 44 . Figure 3 shows the average nearest neighbor degree as a function of the degree k for four selected networks and their values after random rewiring as well as the degree correlation coefficient \(\rho _D\) . Refer to Supplementary Fig. S3 for the ANND plots of all networks. The randomized networks with preserved degree distribution have no degree-degree correlation. As a result, the function ANND does not vary with k . The randomized networks serve as a reference for the expected ANND values when the links are distributed at random.

We find that most semantic networks are disassortative as ANND is a decreasing function of the degree k and the degree correlation coefficient \(\rho _D\) is negative. These networks are ‘Has-A’, ‘Part-Of’, ‘Is-A’, ‘Related-To’ and ‘Union’. In disassortative networks, nodes with larger degrees (general words) tend to connect to nodes with smaller degrees (less general words). This is not surprising. Indeed, if we use these relations in a sentence, then we often relate specific words to more general words. For example, we say ‘horse racing is a sport’, in which ‘horse racing’ is a very specific phrase while ‘sport’ is more general.

On the other hand, network ‘Synonym’ is assortative as the function ANND increases in the degree k . This indicates that large-degree nodes (general words) connect to nodes that have similar degree (words with the same generality). The same applies to network ‘Antonym’. Although the degree correlation is not very pronounced and reflected by the small correlation coefficient \(\rho _D=-0.005\) , we still see a slight upward trend in the curve of ANND.

The function ANND of a rewired network is not degree-dependent anymore, shown by the orange curves in Figure 3 . The curve is almost flat for ‘Synonym’ and ‘Related-To’. At the larger degree k , the curve may drop slightly, as for large-degree nodes there are not enough nodes of equal degree to connect to.

Average nearest neighbor degree (ANND) as a function of degree k and degree correlation coefficient \(\rho _D\) of four English semantic networks. ( a ) Network ‘Has-A’ , ( b ) Network ‘Is-A’ , ( c ) Network ‘Antonym’ , ( d ) Network ‘Synonym’ . See SI for the results of all seven networks. Data points in circles are the average ANND of nodes with degree k in a network, triangles represent the data after logarithmic binning, and squares are the average ANND of nodes with degree k in the rewired network. Logarithmic binning is used to better visualize the data.

Clustering coefficient

In networks such as social networks, the neighbors of a node are likely to be connected as well, a phenomenon which is known as clustering 33 , 53 . If a person has a group of friends, there is a high chance that these friends also know each other. These networks are characterized by many triangular connections.

Figure 4 shows the average clustering coefficient \(c_G(i)\) of nodes with degree \(d_{i}=k\) of four English networks. Refer to Supplementary Fig. S4 for the clustering coefficients of all seven networks. All networks have small clustering coefficients in absolute terms, which, in combination with the small average degree E [ D ], indicates a local tree-like structure. We find that the networks ‘Part-Of’, ‘Antonym’ and ‘Synonym’ have substantially larger clustering coefficients than their rewired counterparts: there are more triangles in these networks than expected by chance. On the other hand, the network ‘Has-A’ has lower clustering coefficients \(c_G(i)\) than the randomized network, therefore it seems that the ‘Has-A’ network is organized in a different way than the other networks. As for the networks ‘Is-A’, ‘Related-To’ and ‘Union’, the clustering coefficients \(c_G(i)\) are similar to their corresponding rewired networks.

In summary, we find that English semantic networks have power-law degree distributions and most are scale-free, which coincides with the results in previous studies 38 , 48 . Besides, semantic networks with different link types show different levels of degree assortativity and average local clustering. Most works in the literature have identified high clustering coefficients in semantic networks 34 , 36 , 38 , 48 . This encourages us to further investigate the organizing principles of these semantic networks, which we will discuss explicitly in the ‘Similarity and complementarity in semantic networks’ section.

The average clustering coefficient \(c_G(i)\) of nodes with degree \(d_{i}=k\) of four English semantic networks. ( a ) Network ‘Has-A’ , ( b ) Network ‘Is-A’ , ( c ) Network ‘Related-To’ , ( d ) Network ‘Synonym’ . See the SI for the results of all seven networks. Data points in circles are the original average local clustering coefficients of nodes with degree \(d_{i}=k\) , triangles represent data after logarithmic binning, and squares show the average clustering coefficients of nodes with degree \(d_{i}=k\) (logarithmically binned) in the randomized networks.

Language-specific properties

Up to this point, we have only considered English semantic networks, while there are thousands of other languages in the world besides English. In this section, we consider semantic networks from 10 other languages contained within ConceptNet: French, Italian, German, Spanish, Russian, Portuguese, Dutch, Japanese, Finnish and Chinese. We group the in total 11 languages based on their language families and we again study the topological properties of 7 different semantic relations per language. Finally, we observe peculiarities in the degree distribution densities of the ‘Related-To’ networks in some languages, which we later explain by grammar.

Language classification

In linguistics, languages can be partitioned in multiple different ways. Mainly, there are two kinds of language classifications: genetic and typological.

The genetic classification assorts languages according to their level of diachronic relatedness, where languages are categorized into the same family if they evolved from the same root language. 54 . An example is the Indo-European family, which includes the Germanic, Balto-Slavic and Italic languages 55 .

Based on these two types of classifications, we have selected 11 languages to cover different language types, Supplementary Table S5 . Typologically, Chinese is an isolating language, while Japanese and Finnish are agglutinating languages. The rest of the languages under consideration (8 out of 11) belong to the inflecting category. Genetically, French, Italian, Spanish and Portuguese belong to the Italic family, while English, German and Dutch are Germanic. Russian is a Balto-Slavic language, Japanese is Transeurasian, Chinese is Sino-Tibetan and Finnish belongs to the Uralic family. We mainly refer to the typological classification throughout our analyses.

Overview of semantic networks from eleven languages

For every language, we construct seven undirected semantic networks with the link types ‘Has-A’, ‘Part-Of’, ‘Is-A’, ‘Related-To’, ‘Union’, ‘Antonym’ and ‘Synonym’. Due to missing data in ConceptNet, only three languages have the ‘Has-A’ relation. For these languages, the ‘Union’ network is the union of three link types: ‘Part-Of’, ‘Is-A’ and ‘Related-To’. In this section, we provide an overview of the numbers of nodes N and numbers of links L of the semantic networks. Again, we restrict our study to the LCCs of these networks.

Regarding the numbers of nodes N , the networks ‘Related-To’ and ‘Union’ are generally the largest networks in a language, with the French ‘Union’ network being the absolute largest with \(N=1,296,622\) , as denoted in Supplementary Table S6 . Nevertheless, there are many small networks with size \(N<100\) , particularly for the ‘Part-Of’ and ‘Synonym’ networks.

Similar to the English semantic networks, we observe that most networks with more than 100 nodes are sparse. All networks have an average degree between 1 and 6, which is small compared to the network size. Consider the Dutch ‘Is-A’ network for example, where a node has about 5 connections on average, which is only 2.45% of 191 nodes in the whole network. Supplementary Table S7 lists the average degree E [ D ] of all our semantic networks.

Many of the semantic networks in the 11 languages have degree distributions that are approximately power laws. We estimate the power-law exponents only for networks with size \(N>1000\) because we require a sufficient number of observations to estimate the power-law exponent \(\gamma\) . Supplementary Table S8 lists the estimated power-law exponents \({\hat{\gamma }}\) using the same 4 methods as in Fig. 2 for each semantic network. Refer to the section ‘Power-law degree distributions’ in SI for details on these estimation procedures.

We find that many networks have power laws in their degree distributions and many of those networks are scale-free ( \(2< {\hat{\gamma }} < 3\) ). The Chinese ‘Related-To’ network even has a power-law exponent \({\hat{\gamma }} <2\) . The degree distributions of all ‘Synonym’ and ‘Antonym’ networks are hardly or not power laws, however. The likely reason for this is that nodes in these networks generally have smaller degrees than in other networks. As a result, the slope of the degree distribution is steeper and therefore not classified as a power law. This is not unexpected, as for a given word there are only a certain number of synonyms or antonyms and therefore there are not many nodes with high degrees. Another interesting finding is that the densities of the degree distributions of the ‘Related-To’ and ‘Union’ networks for French, Spanish, Portuguese and Finnish show notable deviations from a straight line in the log-log plot, which we discuss in-depth in the next section.

Language inflection

In some languages, the densities of the degree distributions of the ‘Related-To’ and ‘Union’ networks show deviations from a straight line on a log-log scale. An example is the Spanish ‘Related-To’ network in Fig. 5 a, where we observe a peak in the tail of the distribution. To find the cause of the anomaly in the degree distribution, we inspect the words with a degree k located in the peak, referred to as peak words , and their neighbors. Supplementary Table S9 lists a few examples of the peak words, which are almost all verbs and have similar spellings. The links adjacent to these nodes with higher-than-expected degrees might be the result of grammatical inflections of the same root words since Spanish is a highly inflected language. We observe a similar anomaly in the degree distributions of French, Portuguese and Finnish ‘Related-To’ and ‘Union’ networks. In Supplementary Table S6 we saw that the network ‘Union’ is mostly composed of ‘Related-To’ in these four languages, therefore we restrict the analysis to the ‘Related-To’ networks.

Two common types of language inflection are conjugation, the inflection of verbs, and declension, the inflection of nouns. The past tense of the verb ‘to sleep’ is ‘slept’, an example of conjugation in English. The plural form of the noun ‘man’ is ‘men’, an example of declension. The languages Spanish, Portuguese and French are much richer in conjugations than English, while Finnish is rich in declensions.

Part-of-speech tags

In the ConceptNet dataset, only part of the nodes is part-of-speech (POS) tagged with one of four types: verb, noun, adjective and adverb. For French, Spanish and Portuguese, the percentage of verbs in the peak is larger than in the LCC, while for Finnish the percentage of nouns in the peak is larger than in the LCC, see Supplementary Table S10 . Remarkably, 100% of the Portuguese peak words are verbs. Most neighbors of the peak words are verbs for Spanish (97%), Portuguese (99%) and French (87%), while most neighbors of the peak words are nouns for Finnish (90%), Supplementary Table S11 . This strengthens our belief that the abnormal number of nodes with a certain degree k is related to language inflection in these four languages.

Merging of word inflections

To investigate whether the peaks in the degree distribution densities are indeed related to word inflections, we leverage the ‘Form-Of’ relation type in ConceptNet, which connects two words A and B if A is an inflected form of B, or B is the root word of A 57 . We merge each node and its neighbors from the ‘Form-Of’ network (its inflected forms) into a single node in the ‘Related-To’ network, as depicted in Supplementary Figure S5 . Figure 5 shows the degree distribution densities of the ‘Related-To’ networks before and after node merging. The range of the anomalous peak in the density of the degree distribution is highlighted in yellow. In each panel, the number of grammatical variations m coincides with the center of the peak. As seen in Fig. 5 a, the peak completely disappears in the Spanish ‘Related-To’ after node merging, thus the peak is described entirely by connections due to word inflections. We also observe significant reductions in the heights of the peaks for Portuguese and Finnish ‘Related-To’ networks. However, for the French ‘Related-To’ network there is only a slight reduction in height after merging, which we believe is likely due to poor coverage in the French ‘Form-Of’ network with only 17% of words in the peak. In contrast, the Spanish ‘Form-Of’ network covers 97% of the Spanish peak words, while for Portuguese and Finnish approximately 50% of the peak words are covered, Supplementary Table S12 .

Densities \(\Pr [D=k]\) of the degree distributions of the ‘Related-To’ networks before and after node merging of inflected forms in ( a ) Spanish, ( b ) French, ( c ) Portuguese and ( d ) Finnish. The logarithmically binned densities after node merging are shown in orange. The peaks are highlighted in yellow. The vertical black lines indicate the number of grammatical variations m for the relevant grammatical rule in the respective language. In each panel the number of grammatical variations m coincides with the center of the peak.

The number of inflections

In a language, the number of distinct conjugations of regular verbs is determined by the number of different pronouns and the number of verb tenses, which are grammatical time references 58 . In Spanish, there are 6 pronouns and 9 simple verb tenses, resulting in at most \(m = 6 \times 9 = 54\) distinct verb conjugations 59 , 60 . Table 1 exemplifies these 54 different conjugations for the verb ‘amar’, which means ‘to love’. There are also irregular verbs that follow different, idiosyncratic grammatical rules, but the majority of the verbs in Spanish are classified as regular, like in most inflecting languages. The number of grammatical variations \(m = 54\) coincides with the center of the peak in Figure 5 a.

Like Spanish, Portuguese has \(m = 54\) distinct conjugated verb forms 61 . In French, there are \(m = 6 \times 7 = 42\) distinct verb conjugations 62 . In Finnish, there are in total 15 noun declensions or cases with distinct spelling, each having singular and plural forms, resulting in \(m = 30\) different cases of a Finnish noun 63 . Supplementary Table S13 lists the number of grammatical variations m in French, Spanish, Portuguese and Finnish, along with the minimum and maximum degree \(k_{min}\) and \(k_{max}\) where the peak starts and ends. By Fig. 5 we confirm that the number of grammatical variations m coincides with or is close to the center of the peak.

In summary, we observe anomalies in the degree distributions of ‘Related-To’ networks from the inflecting languages Spanish, French and Portuguese and the agglutinating Finnish. Because of grammatical structures, root words in these languages share many links with their inflected forms, resulting in more nodes with a certain degree than expected. While Finnish is typologically classified as agglutinating, it still has many noun declensions, suggesting that the agglutinating and inflecting language categories are not mutually exclusive.

Similarity and complementarity in semantic networks

Although we have identified several universal characteristics of semantic networks, we also observe notable differences in some of their properties. The clustering coefficient in some semantic networks, for instance, is greater than expected by chance, while in other semantic networks, e.g. , the English ‘Has-A’ network, the clustering coefficient is smaller than expected by chance.

We hypothesize that these semantic networks are organized according to different principles. It is commonly known that one such principle is similarity: all factors being equal, similar nodes are more likely to be connected. Similarity is believed to play a leading role in the formation of ties in social networks and lies at the heart of many network inference methods. At the same time, recent works indicate that many networks may be organized predominantly according to the complementarity principle, which dictates that interactions are preferentially formed between nodes with complementary properties. Complementarity has been argued to play a significant role in protein-protein interaction networks 64 and production networks 65 . In addition, a geometric complementarity framework for modeling and learning complementarity representations of real networks was recently formulated by Budel and Kitsak 66 .

This section aims to assess the relative roles of complementarity and similarity mechanisms in different semantic networks. We utilize the method by Talaga and Nowak 29 . The method assesses the relative roles of the two principles by measuring the relative densities of triangular and quadrangle motifs in the networks. Intuitively, the transitivity of similarity - A similar to B and B similar to C implies A similar to C – results in a high density of triangles 20 , 67 , 68 , Supplementary Figure S6 a. The non-transitivity of complementarity, on the other hand, suppresses the appearance of triangles but enables the appearance of quadrangles in networks 64 , 69 .

We measure and compare the density of triangles and quadrangles with the structural similarity and complementarity coefficients using the framework of Talaga and Nowak 29 . After computing the densities of triangles and quadrangles, the framework assesses their significance by comparing the densities to those of the configurational models built with matching degree distributions, see the SI for a summary. As a result of the assessment, the network of interest is quantified by two normalized structural coefficients corresponding to complementarity and similarity.

Figure 6 depicts the relative roles of complementarity and similarity in 50 semantic networks. We observe that semantic networks are clustered together according to semantic relation types and not their language families, indicating that specific types of semantic relations matter more for the organizing principles of a semantic network rather than its language.

Calibrated average structural coefficients of the LCCs of the 50 semantic networks from 11 languages. Languages that belong to the same family are marked with similar shapes. Triangles represent Italic, quadrilaterals represent Germanic, circles represent Balto-Slavic, a star represents Transeurasian, a cross represents Sino-Tibetan and a pentagon represents Uralic. The marker size scales logarithmically with the number of nodes N in the network and is further adjusted for visibility. The grey lines at \(x=0\) and \(y=0\) indicate the expected coefficients based on the configuration model (see SI). The dashed line at \(y=x\) indicates that the structural similarity and complementarity coefficients are equal. Networks in the upper left area (shaded in blue) are more complementarity-based, while networks in the lower right area (shaded in yellow) are more similarity-based. We highlight four clusters of networks using different colors.

Based on the calibrated complementarity and similarity values, we can categorize most semantic networks as (i) predominantly complementarity-based, (ii) predominantly similarity-based, and (iii) networks where both complementarity and similarity are substantially present.

We observe four clustering patterns in Figure 6 .

Cluster 1 (light blue): the ‘Synonym’ networks are characterized by stronger similarity than complementarity values. This observation is hardly surprising since ‘Synonym’ networks link words with similar meanings. Since similarity is transitive, the Synonym’ networks contain significant numbers of connected node triples, leading to large clustering coefficients.

Cluster 2 (red): the ‘Antonym’ networks, as observed in Fig. 6 , belong to the upper triangle of the scatter plot plane, hence complementarity is more prevalent in these networks than similarity. This observation is as expected, as antonyms are word pairs with opposite meanings that complement each other. In our earlier work 66 we learned a geometric representation of the English ‘Antonym’ network demonstrating that antonyms indeed complement each other.

More surprisingly, some antonym networks are characterized by substantial similarity values, implying the presence of triangle motifs. This is the case since there are instances of three or more words that have opposite meanings to all other words in the group. One example is the triple of words ( man , woman , girl ). Each pair of words in the triple is opposite in meaning along a certain dimension, here either gender or age.

Cluster 3 (purple): the ‘Has-A’ networks show more complementarity than similarity. Intuitively, words in ‘Has-A’ complement one another. For instance, ‘a house has a roof ’ describes a complementary relation and these two objects are not similar to one another.

In summary, we have conducted an exploratory analysis of the topological properties of semantic networks with 7 distinct semantic relations arising from 11 different languages. We identified both universal and unique characteristics of these networks.

We find that semantic networks are sparse and that many are characterized by a power-law degree distribution. We also find that many semantic networks are scale-free. We observe two different patterns of degree-degree mixing in these networks, some networks are assortative, while some are disassortative. In addition, we find that most networks are more clustered than expected.

On the other hand, some semantic networks – ‘Related-To’ in French, Spanish, Portuguese, and Finnish – have unique features that can be explained by rules of grammatical inflection. Because of the grammar in these languages, words have many conjugations or declensions. We have related anomalous peaks in the degree distributions to the language inflections. Notably, we found word inflection not only in inflecting languages but also in Finnish, which is an agglutinating language.

We have also quantified the relative roles of complementarity and similarity principles in semantic networks. The proportions of similarity and complementarity in networks differ depending on the semantic relation type. For example, the ‘Synonym’ networks exhibit stronger similarity, while the links in the ‘Antonym’ network are primarily driven by complementarity. In addition, the Chinese ‘Related-To’ network has the highest structural complementarity coefficient of all networks, which we attribute to a unique grammatical phenomenon in Chinese: measure words.

Through the analysis of the topological properties of semantic networks, we found that complementarity may play an important role in their formation. Since most of the state-of-the-art network inference methods are either built on or inspired by the similarity principle, we call for a careful re-evaluation of these methods when it comes to inference tasks on semantic networks. One basic example is the prediction of missing links. In a seminal work, Kovács et al. 64 demonstrated that protein interactions should be predicted with complementarity-tailored methods. We expect that similar methods might be in place for semantic networks. Instead of using the triangle closure principle, one might benefit from the methods based on quadrangle closure, Figure 7 .

Examples of similarity and complementarity in semantic networks. a Similarity: triangle closure in the ‘Synonym’ network. b Complementarity: quadrangle closure in the ‘Antonym’ network.

It is not as easy to illustrate quadrangle closure in network embedding methods or NLP methods in general. A plethora of methods use multiple modules and parameters in learning tasks and can, in principle, be better optimized for the complementarity structure of semantic networks. In our recent work, we propose a complementarity learning method and apply it to several networks, including the ‘Antonym’ semantic network 66 .

Recent groundbreaking advances in large language models are attributed to the multi-head attention mechanism of the Transformer, which uses ideas consistent with the complementarity principle 72 . We advocate that a better understanding of statistical mechanisms underlying semantic networks can help us improve NLP methods even further.

We investigate clustering in semantic networks by measuring the clustering coefficient \(c_G(i)\) of a node i , which equals the ratio of the number y of connected neighbors to the maximum possible number of connected neighbors,

as defined by Watts and Strogatz 33 , 42 . The graph clustering coefficient \(c_G\) is the average over all node clustering coefficients,

We also calculate the average clustering coefficient \(c_G(i)\) of nodes with degree \(d_{i}=k\) . In addition, we calculate \(c_G(i)\) also after random rewiring for comparison.

Data Availibility

This study did not generate any new data. Networks used in this study are freely available from https://conceptnet.io .

Cambria, E. & White, B. Jumping NLP curves: A review of natural language processing research. IEEE Comput. Intell. Mag. 9 , 48–57 (2014).

Article Google Scholar

Quillian, M. R. Word concepts: A theory and simulation of some basic semantic capabilities. Behav. Sci. 12 , 410–430 (1967).

Article CAS PubMed Google Scholar

Quillian, M. R. The teachable language comprehender: A simulation program and theory of language. Comm. ACM 12 , 459–476 (1969).

Sowa, J. F. Semantic networks. Encycl. Cognit. Sci. 2 , 1429 (2012).

Google Scholar

de Barros Pereira, H. B. et al. Systematic review of the “semantic network” definitions. Expert Syst. Appl. 118455 (2022).

Sowa, J. F. Principles of semantic networks: Explorations in the representation of knowledge (Morgan Kaufmann, 2014).

Peters, S. & Shrobe, H. Using semantic networks for knowledge representation in an intelligent environment. In Proceedings of the First IEEE International Conference on Pervasive Computing and Communications, 2003 (PerCom 2003) , 323–329, https://doi.org/10.1109/PERCOM.2003.1192756 (2003).

Salem, A.-B. M. & Alfonse, M. Ontology versus semantic networks for medical knowledge representation. Recent Adv. Comput. Eng. 769–774 (2008).

Singhal, A. Introducing the knowledge graph: things, not strings. Google Blog https://blog.google/products/search/introducing-knowledge-graph-things-not (2012).

Popping, R. Knowledge graphs and network text analysis. Soc. Sci. Inf. 42 , 91–106 (2003).

Fensel, D. et al. Introduction: what is a knowledge graph? In Knowledge Graphs , 1–10 (Springer, 2020).

Kepuska, V. & Bohouta, G. Next-generation of virtual personal assistants (Microsoft Cortana, Apple Siri, Amazon Alexa and Google Home). In 2018 IEEE 8th annual computing and communication workshop and conference (CCWC) , 99–103 (IEEE, 2018).

High, R. The era of cognitive systems: An inside look at IBM Watson and how it works. IBM Corp. Redb. 1 , 16 (2012).

Piskorski, J. & Yangarber, R. Information extraction: Past, present and future. In Multi-source, multilingual information extraction and summarization , 23–49 (Springer, 2013).

Shi, F., Chen, L., Han, J. & Childs, P. A data-driven text mining and semantic network analysis for design information retrieval. J. Mech. Des. 139 (2017).

Ouyang, L. et al. Training language models to follow instructions with human feedback. Adv. Neural Inf. Process. Syst. 35 , 27730–27744 (2022).

Resnik, P. Semantic similarity in a taxonomy: An information-based measure and its application to problems of ambiguity in natural language. J. Art. Intell. Res. 11 , 95–130 (1999).

MATH Google Scholar

Harris, A. Human languages vs. programming languages. Medium https://medium.com/@anaharris/human-languages-vs-programming-languages-c89410f13252 (2018).

McPherson, M., Smith-Lovin, L. & Cook, J. M. Birds of a feather: Homophily in social networks. Ann. Rev. Sociol. 415–444 (2001).

Kossinets, G. & Watts, D. J. Origins of homophily in an evolving social network. Am. J. Sociol. 115 , 405–450 (2009).

Schaefer, D. R., Light, J. M., Fabes, R. A., Hanish, L. D. & Martin, C. L. Fundamental principles of network formation among preschool children. Soc. Netw. 32 , 61–71 (2010).

Snijders, T. A. Statistical models for social networks. Ann. Rev. Sociol. 37 , 131–153 (2011).

Hasan, M. A. & Zaki, M. J. A survey of link prediction in social networks. In Social network data analytics , 243–275 (Springer, 2011).

Zarandi, F. D. & Rafsanjani, M. K. Community detection in complex networks using structural similarity. Phys. A: Stat. Mech. 503 , 882–891 (2018).

Evans, N. & Levinson, S. C. The myth of language universals: Language diversity and its importance for cognitive science. Behav. Brain Sci. 32 , 429–448 (2009).

Article PubMed Google Scholar

Khurana, D., Koli, A., Khatter, K. & Singh, S. Natural language processing: State of the art, current trends and challenges. Multimed. Tools. Appl. 1–32 (2022).

Alfawareh, H. M. & Jusoh, S. Resolving ambiguous entity through context knowledge and fuzzy approach. Int. J. Comp. Sci. Eng. 3 , 410–422 (2011).

Jusoh, S. A study on NLP applications and ambiguity problems. J. Theor. Appl. Inf. Technol. 96 (2018).

Talaga, S. & Nowak, A. Structural measures of similarity and complementarity in complex networks. Sci. Rep. 12 , 16580 (2022).

Article ADS CAS PubMed PubMed Central Google Scholar

Nelson, D. L., McEvoy, C. L. & Schreiber, T. A. The University of South Florida word association, rhyme, and word fragment norms. http://w3.usf.edu/FreeAssociation (1998).

Ferrer-i-Cancho, R. & Solé, R. V. The small world of human language. Proc. R. Soc. B: Biol. Sci. 268 , 2261–2265 (2001).

BNC Consortium. British National Corpus, XML edition. http://hdl.handle.net/20.500.12024/2554 (2007). Oxford Text Archive.

Watts, D. J. & Strogatz, S. H. Collective dynamics of ‘small-world’ networks. Nature 393 , 440–442 (1998).

Article ADS CAS PubMed MATH Google Scholar

Motter, A. E., De Moura, A. P., Lai, Y.-C. & Dasgupta, P. Topology of the conceptual network of language. Phys. Rev. E 65 , 065102 (2002).

Article ADS Google Scholar

Ward, G. Moby thesaurus list (Quality Classics, 2015).

Sigman, M. & Cecchi, G. A. Global organization of the WordNet lexicon. Proc. Nat. Acad. Sci. 99 , 1742–1747 (2002).

Miller, G. A. WordNet: A lexical database for english. Comm. ACM 38 , 39–41 (1995).

Steyvers, M. & Tenenbaum, J. B. The large-scale structure of semantic networks: Statistical analyses and a model of semantic growth. Cogn. Sci. 29 , 41–78 (2005).

Roget, P. M. Roget’s Thesaurus of English Words and Phrases (TY Crowell Company, 1911).

Barabási, A.-L. & Albert, R. Emergence of scaling in random networks. Science 286 , 509–512 (1999).

Article ADS MathSciNet PubMed MATH Google Scholar

Strogatz, S. H. Exploring complex networks. Nature 410 , 268–276 (2001).

Van Mieghem, P. Performance Analysis of Complex Networks and Systems (Cambridge University Press, 2014).

Ferrer-i Cancho, R., Solé, R. V. & Köhler, R. Patterns in syntactic dependency networks. Phys. Rev. E 69 , 051915 (2004).

Noldus, R. & Van Mieghem, P. Assortativity in complex networks. J. Complex Netw. 3 , 507–542 (2015).

Article MathSciNet MATH Google Scholar

Speer, R., Chin, J. & Havasi, C. Conceptnet 5.5: An open multilingual graph of general knowledge. In Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence , AAAI’17, 4444–4451 (AAAI Press, 2017).

Voitalov, I., van der Hoorn, P., van der Hofstad, R. & Krioukov, D. Scale-free networks well done. Phys. Rev. Res. 1 , 033034 (2019).

Article CAS Google Scholar

Voitalov, I. Tail index estimation for degree sequences of complex networks. https://github.com/ivanvoitalov/tailestimation (2019).

Borge-Holthoefer, J. & Arenas, A. Semantic networks: Structure and dynamics. Entropy 12 , 1264–1302 (2010).

Article ADS MATH Google Scholar

Petersen, A. M., Tenenbaum, J. N., Havlin, S., Stanley, H. E. & Perc, M. Languages cool as they expand: Allometric scaling and the decreasing need for new words. Sci. Rep. 2 , 943 (2012).

Article ADS PubMed PubMed Central Google Scholar

Adamic, L. A. & Huberman, B. A. Power-law distribution of the world wide web. Science 287 , 2115–2115 (2000).

Jeong, H., Tombor, B., Albert, R., Oltvai, Z. N. & Barabási, A.-L. The large-scale organization of metabolic networks. Nature 407 , 651–654 (2000).

Article ADS CAS PubMed Google Scholar

Faloutsos, M., Faloutsos, P. & Faloutsos, C. On power-law relationships of the internet topology. ACM SIGCOMM Comp. Commun. Rev. 29 , 251–262 (1999).

Article MATH Google Scholar

Newman, M. E. Clustering and preferential attachment in growing networks. Phys. Rev. E 64 , 025102 (2001).

Article ADS CAS Google Scholar

Lyons, J. Language classification. Encyclopædia Britannica https://www.britannica.com/science/linguistics/Other-relationships (2022).

Eberhard, D. M., Simons, G. F. & Fennig, C. D. Ethnologue: Languages of the world. SIL International https://www.ethnologue.com (2022).

Haspelmath, M. The morph as a minimal linguistic form. Morphology 30 , 117–134 (2020).

Speer, R. Relations in ConceptNet5. ConceptNet 5 Wiki https://github.com/commonsense/conceptnet5/wiki/Relations (2019).

Comrie, B. Aspect: An introduction to the study of verbal aspect and related problems Vol. 2 (Cambridge University Press, 1976).

Kendris, C. & Kendris, T. 501 Spanish verbs (Barrons Educational Series, 2020).

Vare, F. J. Your all-in-one guide to the 18 Spanish tenses and moods. Enux Education, FluentU https://www.fluentu.com/blog/spanish/spanish-tenses/ (2022).

Nitti, J. J. & Ferreira, M. J. 501 Portuguese verbs (Simon and Schuster, 2015).

Lawless, L. K. The Everything French Verb Book: A Handy Reference for Mastering Verb Conjugation (Simon and Schuster, 2005).

Karlsson, F. Finnish: A comprehensive grammar (Routledge, 2017).

Kovács, I. A. et al. Network-based prediction of protein interactions. Nat. Commun. 10 , 1–8 (2019).

Mattsson, C. E. et al. Functional structure in production networks. Front. Big Data 4 , 666712 (2021).

Article PubMed PubMed Central Google Scholar

Budel, G. & Kitsak, M. Complementarity in complex networks. arXiv preprint arXiv:2003.06665 (2020).

Rivera, M. T., Soderstrom, S. B. & Uzzi, B. Dynamics of dyads in social networks: Assortative, relational, and proximity mechanisms. Ann. Rev. Sociol. 36 , 91–115 (2010).

Asikainen, A., Iñiguez, G., Ureña-Carrión, J., Kaski, K. & Kivelä, M. Cumulative effects of triadic closure and homophily in social networks. Sci. Adv. 6 , eeax7310 (2020).

Jia, M., Gabrys, B. & Musial, K. Measuring quadrangle formation in complex networks. IEEE Trans. Netw. Sci. Eng. 9 , 538–551 (2021).

Article MathSciNet Google Scholar

Tai, J. H. Chinese classifier systems and human categorization. In honor of William S.-Y. Wang Interdisciplinary studies on language and language change (Pyramid Press Taipei, 1994).

Cheng, L.L.-S. & Sybesma, R. Yi-wan tang, yi-ge tang: Classifiers and massifiers. Tsing Hua J. Chin. Stud. 28 , 385–412 (1998).

Vaswani, A. et al. Attention is all you need. Adv. Neural Inf. Process. Syst. 30 (2017).

Download references

Acknowledgements

We thank R. Kooij for useful discussions and suggestions.

This work is part of NExTWORKx, a collaboration between TU Delft and KPN on future telecommunication networks. Parts of this research have been funded by the European Research Council under the European Union’s Horizon 2020 research and innovation program (Grant Agreement 101019718) and the Dutch Research Council (NWO) grant OCENW.M20.244.

Author information

These authors contributed equally: Gabriel Budel and Ying Jin.

Authors and Affiliations

Faculty of Electrical Engineering, Mathematics and Computer Science, Delft University of Technology, 2628 CD, Delft, The Netherlands

Gabriel Budel, Ying Jin, Piet Van Mieghem & Maksim Kitsak

You can also search for this author in PubMed Google Scholar

Contributions

M.K. conceived the experiments, Y.J. conducted the experiments with support from G.B. and M.K, after which G.B., Y.J., P.V.M. and M.K. analyzed the results. All authors wrote, edited, and reviewed the manuscript.

Corresponding author

Correspondence to Maksim Kitsak .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Supplementary information., rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Budel, G., Jin, Y., Van Mieghem, P. et al. Topological properties and organizing principles of semantic networks. Sci Rep 13 , 11728 (2023). https://doi.org/10.1038/s41598-023-37294-8

Download citation

Received : 24 April 2023

Accepted : 19 June 2023

Published : 20 July 2023

DOI : https://doi.org/10.1038/s41598-023-37294-8

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

Explore articles by subject
Guide to authors
Editorial policies

Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.

Breadcrumbs Section. Click here to navigate to respective pages.

Semantic Network Analysis in Social Sciences

DOI link for Semantic Network Analysis in Social Sciences

Get Citation

Semantic Network Analysis in Social Sciences introduces the fundamentals of semantic network analysis and its applications in the social sciences. Readers learn how to easily transform any given text into a visual network of words co-occurring together, a process that allows mapping the main themes appearing in the text and revealing its main narratives and biases.

Semantic network analysis is particularly useful today with the increasing volumes of text-based information available. It is one of the developing, cutting-edge methods to organize, identify patterns and structures, and understand the meanings of our information society. The first chapters in this book offer step-by-step guidelines for conducting semantic network analysis, including choosing and preparing the text, selecting desired words, constructing the networks, and interpreting their meanings. Free software tools and code are also presented. The rest of the book displays state-of-the-art studies from around the world that apply this method to explore news, political speeches, social media content, and even to organize interview transcripts and literature reviews.

Aimed at scholars with no previous knowledge in the field, this book can be used as a main or a supplementary textbook for general courses on research methods or network analysis courses, as well as a starting point to conduct your own content analysis of large texts.

Chapter | 15 pages, introduction, chapter 1 | 16 pages, how to conduct semantic network analysis, chapter 2 | 21 pages, the news coverage of threats, chapter 3 | 19 pages, provocation narratives in chinese and us newspapers, chapter 4 | 22 pages, cable news channels' partisan ideology and market share growth as predictors of social distancing sentiment during the covid-19 pandemic, chapter 5 | 18 pages, politicizing the holocaust, chapter 6 | 24 pages, network of cleavages, chapter 7 | 23 pages, sexual assaults blindsided by politics on twitter, chapter 8 | 16 pages, time to be happy, chapter 9 | 17 pages, school improvement, chapter 10 | 24 pages, identifying patterns in communication science, chapter 11 | 12 pages, summary and conclusion, chapter | 3 pages.

Privacy Policy
Terms & Conditions
Cookie Policy
Taylor & Francis Online
Taylor & Francis Group
Students/Researchers
Librarians/Institutions

Connect with us

Corpus ID: 220267426

What Constitutes Semantic Network Analysis? A Comparison of Research and Methodologies'

Marya L. Doerfel
Published 2003
Computer Science, Linguistics

Tables from this paper

158 Citations

Bridging semantic and social network analyses: the case of the hashtag #precisamosfalarsobreaborto (we need to talk about abortion) on twitter.

Highly Influenced

The Pivotal Role of AJC in the Growth of Communication Research in Asia: A semantic network analysis

Identifying policy frames through semantic network analysis: an examination of nuclear energy policy across six countries, the motivation of social entrepreneurs: the roles, agendas and relations of altruistic economic actors, semantics of online tourism and travel information search on the internet: a preliminary study, relaciones, redes y discurso, a data-driven approach for discovery of the latest research trends in higher education for business by leveraging advanced technology and big data, the latest research trends of teaching & learning in higher education by leveraging advanced technology and big data, a new approach to semantic sustainability assessment: text mining via network analysis revealing transition patterns in german municipal climate action plans, institutional hybridity in public sector reform: replacement, blending, or layering of administrative paradigms, 26 references, emergent communication networks.

Highly Influential

A Semantic Network Analysis of the International Communication Association

Cultural differences in organizational communication: a semantic network analysis 1, european managers' interpretations of participation: a semantic network analysis., is it really just like a fancy answering machine comparing semantic networks of different types of voice mail users, the structure of communication: a network analysis of the international communication association., friendship patterns and culture: the control of organizational diversity, triangulation as a research strategy for identifying invisible colleges among biomedical scientists, semantic memory: facts and models., related papers.

Showing 1 through 3 of 0 Related Papers

Semantic network analysis in consumer and marketing research: application areas in phygital contexts

Qualitative Market Research

ISSN : 1352-2752

Article publication date: 21 May 2024

Issue publication date: 7 June 2024

Large-scale text-based data increasingly poses methodological challenges due to its size, scope and nature, requiring sophisticated methods for managing, visualizing, analyzing and interpreting such data. This paper aims to propose semantic network analysis (SemNA) as one possible solution to these challenges, showcasing its potential for consumer and marketing researchers through three application areas in phygital contexts.

Design/methodology/approach

This paper outlines three general application areas for SemNA in phygital contexts and presents specific use cases, data collection methodologies, analyses, findings and discussions for each application area.

The paper uncovers three application areas and use cases where SemNA holds promise for providing valuable insights and driving further adoption of the method: (1) Investigating phygital experiences and consumption phenomena; (2) Exploring phygital consumer and market discourse, trends and practices; and (3) Capturing phygital social constructs.

Research limitations/implications

The limitations section highlights the specific challenges of the qualitative, interpretivist approach to SemNA, along with general methodological constraints.

Practical implications

Practical implications highlight SemNA as a pragmatic tool for managers to analyze and visualize company-/brand-related data, supporting strategic decision-making in physical, digital and phygital spaces.

Originality/value

This paper contributes to the expanding body of computational, tool-based methods by providing an overview of application areas for the qualitative, interpretivist approach to SemNA in consumer and marketing research. It emphasizes the diversity of research contexts and data, where the boundaries between physical and digital spaces have become increasingly intertwined with physical and digital elements closely integrated – a phenomenon known as phygital.

Data visualization
Semantic network analysis
Automated text analysis
Exploratory data analysis
Computational text analysis

Schöps, J.D. and Jaufenthaler, P. (2024), "Semantic network analysis in consumer and marketing research: application areas in phygital contexts", Qualitative Market Research , Vol. 27 No. 3, pp. 495-514. https://doi.org/10.1108/QMR-06-2023-0084

Emerald Publishing Limited

Published by Emerald Publishing Limited. This article is published under the Creative Commons Attribution (CC BY 4.0) licence. Anyone may reproduce, distribute, translate and create derivative works of this article (for both commercial & non-commercial purposes), subject to full attribution to the original publication and authors. The full terms of this licence may be seen at http://creativecommons.org/licences/by/4.0/legalcode

Introduction

The analysis of language plays a pivotal role in consumer and marketing research ( Packard and Berger, 2023 ). Traditionally, researchers have used various text-based methods, such as interviews, open-ended survey questions and association tasks, to gain a deeper understanding of consumers. The emergence of social media platforms, big data, and the appification of services has not only profoundly expanded the realm of text-based consumer-generated data ( Klostermann et al. , 2018 ) but also blurred the boundaries between physical and digital environments ( Batat, 2022 ). This blurring has given rise to environments that integrate “physical, human, digital, and media content elements, platforms, technologies, and extended realities” ( Batat, 2022 : 10) – phygital environments. In this confluence of the physical and digital, social media serves as a bridge, enabling users to translate their real-world experiences onto digital platforms ( Schöps et al. , 2024 ), thus creating complex networks of phygital interactions. User-generated content on social media introduces a digital layer to physical experiences, as these shared narratives and visual stories infuse digital spaces with elements of the tangible world. Conversely, the digital feedback loop influences physical experiences, as real-world actions, representations and choices are often shaped by digital interactions, trends and community engagements ( Beichert et al. , 2024 ), further blurring the lines between these realms.

Traditional data analysis methods, such as manual content analysis, often struggle to account for the intertwined nature of physical and digital elements, as well as their size and scope. It is, therefore, unsurprising that computational methods, such as automated text analysis (ATA) and semantic network analysis (SemNA), have gained notable attention recently. These methods facilitate the analysis of the nature, size and scope of unstructured, large-scale text-based data. ATA ( Berger et al. , 2020 , 2022 ) quantifies unstructured, large-scale text-based data, enabling researchers to test “the relationship between or amongst the constructs or variables of interest” ( Humphreys and Wang, 2018 : 1275). SemNA, a related yet distinct approach, centers on the visualization, exploration and analysis of “the web of meaning” ( Doerfel, 1998 : 17) or the “networks of interrelated conceptual units” ( Lehmann, 1992 : 1) in text-based data. This method facilitates the integration of both quantitative elements, such as numeric measures of semantic units and their co-occurrences, and qualitative components, involving the visualization and interpretation of the relationships in semantic networks ( Rice, 2005 ).

In phygital contexts, exploring, understanding and accounting for the nature, scope and relationships between physical and digital elements is particularly crucial. This study introduces SemNA as both a tool and a solution for researchers navigating these complex phygital domains of consumer and marketing research. Despite existing work in related fields that outlines the procedural steps for SemNA ( Christensen and Kenett, 2021 ; Segev, 2021 ) – enabling its replicability in other fields – a notable gap exists in guidance on when , where and how to apply it in consumer and marketing research unfolding in contemporary phygital contexts. This lack of detailed guidance is a major barrier to the method’s adoption and effective utilization in the field. Accordingly, identifying and addressing these areas are essential steps toward broadening the application and impact of SemNA in consumer and marketing research.

This paper proposes three application areas, accompanied by exemplary use cases, for SemNA within phygital contexts, targeting consumer and marketing research. The first application area uses SemNA to explore the complex relationships between digital and physical consumption practices. This is illustrated by an analysis of narrative interviews in which respondents were immersed in the digital space of Instagram, serving as a stimulus for reflective accounts of the interplay between digital fashion content consumption and its impact on physical fashion consumption. The second application area uses SemNA to investigate the phygital nature of social media discourse, trends and practices. This is illustrated through an examination of the #sustainability discourse, revealing its manifestation as a product of both digital affordances and physical consumption practices. The third application area uses SemNA to convert social constructs of phygital nature into tangible and intuitive digital representations. This is exemplified by illustrating how mental associations with Cristiano Ronaldo are composed of his physical achievements and his social media persona. Accordingly, this paper contributes to “the growing need for methodologies that incorporate analysis of textual data in consumer [and marketing] research” ( Humphreys and Wang, 2018 : 1296). We provide an overview of each application area encompassing purposes, benefits, challenges, exemplary research questions and data sources, thereby showcasing its potential and versatile application across textual data in phygital contexts.

Theoretical background

Computational text analysis methods in consumer and marketing research.

Computational text analysis methods are increasingly used in consumer and marketing research for large-scale text analysis and natural language processing. These methods encompass, for example, dictionary-based text analysis, word embedding models, topic modeling ( Berger et al. , 2022 ) and SemNA ( Caliandro and Gandini, 2017 ), which differ in their approaches and outputs. Dictionary-based text analysis relies on predefined categories or dictionaries. Linguistic inquiry and word count, for example, categorizes words into various emotional, cognitive and structural components and outputs a quantitative analysis showing the frequency or presence of certain categories of words ( Berger et al. , 2022 ). Word embedding models “learn semantic representations from word co-occurrence patterns in natural language” ( Berger et al. , 2022 : 370), and output a set of vector representations for words. Instead of explicitly mapping relationships between concepts, these models encode relationships within a multidimensional space, where each word is represented as a vector. Topic modeling is a statistical method that identifies topics present in a corpus of text and outputs a set of topics, each represented as a collection of terms and a distribution of these topics within each document. As such, topic modeling provides statistical insights into the prevalent themes in the text ( Berger et al. , 2022 ) but does not capture the explicit relationships between words. In contrast, SemNA focuses on visualizing “set[s] of connected semantic relations” – paths – that “can be analyzed qualitatively to reveal different semantic relations that connect words in the semantic network” ( Drieger, 2013 : 7, 10). This approach enables a more nuanced exploration of the relationships and representations of concepts within textual data.

Key elements of semantic network analysis

SemNA has been developed across multiple disciplines over the past few decades ( Danowski, 1988 ). While early accounts focused on the ontology and grammar of words, the paradigmatic focus shifted from the “grammatical function of words” to “the study of the relationships between words, concepts, and meanings” ( Segev, 2021 : 5). As such, SemNA focuses on the visualization and analysis of “meaning networks” ( Doerfel, 1998 : 24), assessing “the extent to which words are related, which indicates something about their meaning” by “the extent to which word pairs co-occur within a given meaning unit” ( Rice, 2005 : 288). This “underlying co-occurrence metric helps to identify groups of [words] that are strongly connected and therefore represent underlying topics” ( Klostermann et al. , 2018 : 553), which can then be explored in-depth through subsequent qualitative, interpretivist work.

Modularity classes.

Nodes, depicted as circles, serve to “represent semantic or lexical units” ( Christensen and Kenett, 2021 : 860). The nature of these semantic units within the network is task-dependent, encompassing distinct concepts or entities. Specifically, these units can be words, phrases or other identifiable meaning-bearing units, such as “category exemplars (verbal fluency), associations to cue words (free association), or cue words whose similarities are rated (similarity judgments)” ( Christensen and Kenett, 2021 : 860). Edges, represented as lines, indicate “the similarity, co-occurrence, or strength of the associations between them [nodes]” ( Christensen and Kenett, 2021 : 860). When analyzing relationships among semantic units, it is important to define the window size, representing the maximum distance within which two semantic units are considered to co-occur within a text (e.g. a window size at the sentence level only connects semantic units within each sentence). A node’s degree – its centrality in the network – is calculated by the total number of edges connected to it. The third central element, modularity classes, is based on an algorithm that iteratively groups nodes into thematic communities in a way that maximizes the modularity of the partition, i.e. the strength of the division of the network into communities ( Blondel et al. , 2008 ). This strength is determined by the modularity score (range: 0 to 1). The higher the value, the more distinct and well-defined these communities are, i.e. the stronger the community structure ( Brandes et al. , 2008 ). When calculating modularity, the resolution parameter, which influences the sensitivity of the algorithm to community detection, can be adjusted. A high resolution favors detecting smaller, and a low resolution larger communities. Notably, no universally correct resolution value exists; it needs to be adjusted to the nature of the data ( Bruns and Snee, 2022 ). Table 1 provides an overview of these three central elements, including their descriptions, measurements and visual representations.

Approaches to semantic network analysis in consumer and marketing research

In consumer and marketing research, two approaches to SemNA have recently developed – a quantitative and a qualitative, interpretivist approach. Both approaches share common ground in their use of SemNA to investigate and analyze the complex nature of unstructured, large-scale text-based data, transforming it into structured analysis that reveals the nuanced interplay of themes, patterns and relationships. The quantitative approach systematically uses text-mining techniques and network analysis tools to measure structural characteristics of semantic networks, such as graph density, centrality measures and clustering coefficients. This approach enables the extraction of quantifiable information from large data sets, facilitating rigorous, replicable analyses that offer insights into phenomena like market structures and consumer behaviors. Studies such as those by Teichert and Schöntag (2010) , Netzer et al. (2012) and Ko et al. (2015) exemplify this approach by quantifying perceptual associations and exploring relationships between concepts such as shopping and mood alleviation, respectively.

Conversely, the qualitative, interpretivist approach integrates SemNA as a tool for initial data structuring, followed by data-driven, interpretivist exploration and theorization ( Lucarelli et al. , 2023 ) of thematic clusters and relationships, delving into the sociocultural intricacies and dynamics revealed by the network analysis ( Caliandro and Gandini, 2017 ) – largely within digital and phygital contexts. For example, Schöps et al. (2020 , 2022 , 2024 ) use SemNA to visualize and interpret semantic discourses on Instagram, exploring how hashtags influence consumer discourse and the structure of digital market networks.

In this paper, we focus on the qualitative, interpretivist approach, highlighting SemNA as a promising method for consumer and marketing researchers to explore various phenomena across phygital contexts ( Batat, 2022 ). Phygital contexts, characterized by the interplay between digital and physical interactions ( Mele and Russo-Spena, 2022 ), demand tools capable of grasping this interplay, generating insights, for example, into the relationship between physical experiences and online discourses. Moreover, the complexity and intertwined nature of data in phygital contexts require methods that can be adapted to diverse text-based data sets and incorporate both quantitative and qualitative elements to enable sophisticated analysis and in-depth interpretations. As SemNA enables the visual mapping of semantic units through computational text analysis, revealing word/concept relationships, thematic clusters and thereby facilitating the qualitative exploration of knowledge, information, concepts and meanings in text-based data, we propose it as a method particularly suited for phygital contexts.

Application areas for semantic network analysis in phygital contexts

Investigating phygital experiences and consumption phenomena;

Exploring phygital consumer and market discourse, trends and practices; and

Capturing phygital social constructs.

The selection of these application areas stems from their direct relevance to the current challenges in understanding consumer behavior within rapidly evolving phygital landscapes. Each area represents a critical facet of consumer and marketing research, where SemNA can offer valuable insights into the complex interplay of digital and physical realms. Through exemplary use cases for each area, we demonstrate the versatility of SemNA and its potential for qualitative, interpretivist consumer and marketing research in phygital contexts. Additionally, Figure 1 provides an overview of the proposed application areas for SemNA in consumer and marketing research, detailing its distinct purposes, benefits, challenges and offering exemplary research questions and data sources for future research.

Application area 1: investigating phygital experiences and consumption phenomena

Contemporary consumption phenomena are increasingly unfolding at the intersection of physical and digital realms, with social media acting as a critical mediator that connects these two spaces. This dynamic interplay renders consumption experiences inherently phygital – the digital sharing of personal experiences on platforms such as Instagram intersects with, and profoundly influences, physical consumption ( Batat, 2022 ). Within this context, investigating consumer narratives of phygital experiences and consumption offers rich insights into their behaviors, preferences and identities. Traditional qualitative data analysis techniques and even modern coding software such as NVivo, which outputs word counts and weighted word percentages, fall short in uncovering the Web of interactions and influences between physical and digital consumption spaces inherent to such phygital consumption phenomena. In this first application area, we showcase how SemNA transcends these limitations by providing a more nuanced understanding of phygital experiences and consumption practices through the visualization of the prominence and interconnectedness – i.e. word pair centralities and relationships – of concepts in phygital data.

As a use case, we chose to investigate consumers’ identity projects related to fashion consumption on Instagram. This use case explores the phygital interplay between digital content consumption on Instagram, e.g. influencer-driven advertorials and its impact on consumers’ physical fashion consumption and identity evolution.

Sample and procedure.

The role of fashion and Instagram in interviewees’ everyday life;

The selected accounts (projective material); and

The role of Instagram in interviewees’ own fashion consumption.

We used a purposive sampling strategy, requiring interviewees to not only have an Instagram account but also to actively follow fashion-related accounts. Interviewees were asked to select two fashion-related accounts of their choice, which served as projective material. The sample comprised 15 European interviewees ( M age = 24 years; 80% female). The interviews lasted from 40 min to 1h 40 resulting in 174 pages of verbatim transcripts.

To prepare the transcripts for subsequent ATA and SemNA, we removed all interviewer questions. We then used Wordij 3.0 ( Danowski, 2013a ) to conduct an ATA. The settings for the ATA included a stop word list ( Humphreys and Wang, 2018 ), which eliminated words such as “I,” “you,” “because” and “example” from the data set. Additionally, words and word pairs that occurred fewer than two times were excluded, word pair order was not preserved, the window size was set to 10 at an intersentential level ( Danowski, 2013b ), English contracted forms were expanded and numbers and punctuation within words were removed.

This procedure resulted in 858 unique words (nodes) and 10,576 connections (edges) between the words. We then used Gephi ( Bastian et al. , 2009 ) to calculate the degree, and run the modularity algorithm (resolution: 1.0; modularity score: 0.206) for community detection ( Brandes et al. , 2008 ). To improve readability, we filtered out words with a degree below 10, resulting in visibility for 58.16% of nodes (499 words) and 80.56% of edges ( Figure 2 ).

Community detection identified three clusters – the purple cluster accounting for 38.23%, the orange cluster for 31.12% and the green cluster for 30.65%. The purple cluster contains narratives about the role of fashion and Instagram in interviewees’ fashion consumption. The words with the highest degree in this cluster are “fashion” (416), “style” (288), “clothes” (275), “wear” (270), “brands” (261), “nice” (250), “buy” (212) and “trends” (180) – core aspects of interviewees’ fashion discourse. In-depth analysis of this cluster further reveals a dense network of connections among the words “fashion” (416), “style” (288), “young” (28), “older” (13), “age” (18) and “change/d” (117), forming paths of relationships that connect the concepts of “fashion” and “identity.” The data reveals various narrative paths constructed by consumers to articulate the transformation of their identities and fashion preferences over time. One such path exemplifying the interconnectedness of these concepts is represented by the following quote:

I’ve changed since I’m studying, going to university. My style changed a lot. Before, in school, it was really different. And I think with growing older, you know more who you are. You find yourself, and maybe are more okay with yourself. Because when you’re younger (…), I think you’re struggling sometimes. (…) you do not know your identity that well.

The orange cluster encapsulates narratives about the role of Instagram in interviewees’ daily interactions with the platform. The words with the highest degree in this cluster are “instagram” (344), “influencers” (290), “people” (273), “day” (252), “stories” (221), “friends” (214), “follow” (201), “videos” (142) and “content” (131). SemNA further uncovers a network of connected semantic relationships between “instagram” (344), “day” (252), “work” (111), “morning” (48), “bored” (43), “evening” (28), “bed” (26), “break” (20) and “routine” (13), forming paths that illuminate how daily digital routines on Instagram seamlessly intersect with and shape physical activities, from work habits to leisure, showcasing the phygital nature of contemporary lifestyle integration. The following quote exemplifies this:

A typical situation would be in the morning after I wake up. Then, a lot of times I use it when I’m at work (…). In my lunch break (…), in my coffee break. After work, when I’m cooking, (…) when I’m eating dinner, and before I go to bed, so all day long. But typical situations would be when I don’t, um, exactly have like when I’m not in a rush.

The green cluster encompasses narratives about the projective material. The words with the highest degree in this cluster are “accounts” (327), “posts” (314), “outfits” (294), “pictures” (279), “life” (260), “good” (204), “person” (171), “cool” (143) and “post” (103). A deeper analysis of the semantic network reveals paths interconnecting the words “accounts” (327), “posts” (314), “pictures” (279), “outfits” (294), “life” (260), “authentic” (51), “perfect” (48), “private” (28) and “honest” (24), illustrating the words and concepts that interviewees relate when talking about digital content and creators, and the qualities they find relatable:

I like her because she appears to be authentic and she puts a lot of work into her posts, which I really appreciate. Also, unlike many other influencers, her accounts only focus on fashion, which means she keeps her private life mostly out of it. Additionally, her style heavily resembles mine, so I can really identify with her.

However, our analysis also reveals that interviewees connect these terms to the potential negative impacts of digital content consumption on the perception of physical reality. The following quote, while initially anchored to the same node, i.e. “authentic,” diverges in its narrative path. That is, the first quote forms a path of relationship between “authentic” and “identity”; the following one between “authentic” and “reality.” This contrast underscores how the relationships between specific words, and consequently their meanings, can vary substantially among individuals. The following quote exemplifies this variation:

(…) when I’m saying that they are authentic and honest and things like that, I know that they are trying to present themselves from their best side. It is also a dangerous platform, especially for younger people, because there is often the urge to compare yourself with those influencers and ask yourself why you don’t have a life like them. So, you shouldn’t lose reality and know that their life is not always as perfect as they portray it to be.

This connection between digital content’s perceived authenticity and consumers’ perception of physical reality underscores a phygital interaction, where digital narratives directly affect offline self-concept and fashion consumption practices. This dynamic also illustrates a considerable phygital challenge, highlighting how the digital world’s pressures and ideals can affect individuals’ self-esteem and consumption practices in the physical world. Accordingly, the complex interplay between digital engagement and physical behaviors reveals the profound influence of social media narratives on personal identity and consumption lifestyles.

Discussion application area 1.

This use case demonstrates the suitability of SemNA in the study of phygital phenomena. Its ability to identify and visualize thematic clusters and the relationships between them provides a nuanced lens through which the complex interplay between digital content consumption and physical consumption practices – in our case, digital fashion content and related offline fashion consumption – becomes deeply insightful. The specific insights gained offer a holistic overview of phygital consumer journeys and the broader implications of phygital experiences and consumption phenomena, showcasing how digital consumption is linked to and impacts offline consumption. SemNA thus facilitates the unraveling of the complexities of phygital interactions which can substantially enrich subsequent interpretivist work ( Lucarelli et al. , 2023 ).

Application area 2: exploring phygital consumer and market discourse, trend and practices

Digital environments are teeming with diverse text-based data, where language is platform-specific. Each platform “comes to have its own unique combination of styles, grammars, and logics, which can be considered as constituting a ‘platform vernacular’” ( Gibbs et al. , 2015 : 257). This vernacular is shaped by platforms’ affordances ( Caliandro and Anselmi, 2021 ). A prevalent text-based affordance in digital environments is the hashtag, which users leverage to frame discursive spaces – referred to as issue spaces ( Marres, 2015 ) – using a specific issue language, such as #blacklivesmatter or #climatechange. Essentially, this process transfers issues from the physical to the digital world, rendering these issues phygital. In marketing contexts, hashtags help frame a “market-embedded and market-related social issue into platform jargon” ( Schöps et al. , 2022 : 79). However, given the vast volume and complex nature of digital data, even specialized qualitative methods like netnography ( Kozinets, 2020 ), developed for social media research, may not fully capture the breadth and depth of digital interactions and the related cultural dynamics. This oversight can leave critical patterns and trends in the data unexplored, which could otherwise enrich subsequent interpretivist work. In this second application area, we demonstrate how SemNA addresses these limitations by visualizing the relationships within the language of these phygital market spaces.

As a use case, we chose to illustrate consumer-generated discourse, trends and practices within the issue space of #sustainability on Instagram. Our analysis focuses on hashtags that co-occur with #sustainability in consumer captions. Sustainability is a contemporary hot topic with pervasive relevance across societies and industries.

We collected a data set comprising metadata from 1,000 Instagram posts tagged with “#sustainability” in February 2023, using the InstaCrawlR scripts in RStudio for targeted data extraction ( Schröder, 2018 ). The metadata includes elements such as post URLs and captions with hashtags. Using a custom Python script, we set the window size at the post level to pair all hashtags within each post, and then compiled this data into a comma separated values file for analysis. This file was subsequently imported into Gephi ( Bastian et al. , 2009 ; Gephi, 2024 ), where we calculated network statistics and visualized the semantic network of hashtags. Specifically, we calculated the degree and ran the modularity algorithm (resolution: 1.0; modularity score: 0.636) for community detection ( Brandes et al. , 2008 ). The resulting network contains 7,656 nodes, representing unique hashtags and 114,202 edges. To improve readability, we filtered out hashtags with a degree below 40, resulting in visibility for 11.79% of nodes and 17.4% of edges ( Figure 3 ).

SemNA identified nine distinct clusters of densely connected hashtags, each representing a unique aspect of the #sustainability discourse. The purple cluster (15.14% of the semantic network) is characterized by general sustainability buzzwords, e.g. #sustainable (degree: 1,855), #ecofriendly (1,635) and #sustainableliving (1,540). The red cluster (7.77%) focuses on sustainable nutrition and food culture, featuring hashtags like #organic (449) and #plantbased (300). The dark gray cluster (6.66%) features hashtags that illustrate consumption practices related to #sustainablefashion (1,309), such as buying #vintage (375), #secondhand (207) and #thriftedfashion (163). The orange cluster (5.57%) is centered around sustainable approaches to #architecture (292) and #interiordesign (404). The blue cluster (4.94%) relates to the #slowfashionmovement (174) seeking a #fashionrevolution (193). The turquoise cluster (4.09%) contains content showcasing #handmade (606) #craftsmanship (73). In the light blue cluster (4%), consumers campaign for #shoplocal (513) and #supportsmallbusinesses (258). The yellow cluster (4.04%) exhibits content related to a #healthylifestyle (186) and #selfcare (175). The light green cluster (2.98%) is focused on #sustainabledevelopmentgoals (93), the green cluster (2.95%) on sustainable #travel (102) content and the pink cluster (2.61%) on #renewableenergy (158), completing the thematic clusters.

Analyzing the relations between semantic units across clusters enables the interpretation of the network’s phygital nature. For example, SemNA reveals connections between #slowfashion, #shoplocal, #supportsmallbusinesses, #craftsmanship, #artisan and #handmade, illustrating a shift toward more sustainable and ethical consumption patterns. By prioritizing slow fashion, consumers advocate moving away from the environmentally harmful and ethically questionable practices of the fast fashion industry toward more sustainably and ethically produced clothing. The emphasis on shopping locally and supporting small businesses highlights a commitment to reducing the carbon footprint from long-distance goods transportation and fostering local economies. Moreover, the focus on craftsmanship and artisanal, handmade products underscores an appreciation for the quality, uniqueness and story behind each item, contrasting with the homogeneity and disposability of mass-produced goods. Together, these connections shed light on consumers’ physical consumption practices, which are then shared in the digital environment of Instagram to create awareness and campaign for sustainable consumption. This sharing and promoting of sustainable choices on Instagram exemplifies how physical experiences and consumption practices intertwine with digital activism.

Discussion application area 2.

This use case highlights SemNA’s utility in exploring the phygital nature of social media discourse, trends and practices – in our case, #sustainability on Instagram. It demonstrates how SemNA provides a comprehensive overview of the intertwined physical and digital elements that constitute social media discourse. This includes a discursive phygital space characterized by digital displays of physical consumption practices enhanced with digital affordances such as hashtags. This holistic visualization not only uncovers the digital language of the specific discourse but also illuminates related trends and practices in the physical world, as illustrated by the interpretive analysis of the cross-cluster connections. Therefore, this use case showcases how SemNA’s capability to produce insightful visualizations that support subsequent interpretivist work ( Lucarelli et al. , 2023 ) on phygital consumer and market discourse, trends and practices, such as “changing and emerging market understandings, and […] shifts in consumer behavior” ( Schöps et al. , 2022 : 98).

Application area 3: capturing phygital social constructs

In physical environments, social constructs materialize in people’s minds. Researchers have continuously sought to better understand and capture such constructs – abstract concepts without objective meaning that have “been created and accepted by the people in a society” ( Merriam-Webster Dictionary, 2024 ). For example, the mental representation of a brand can be conceptualized as compositional associations triggered when consumers think of the brand ( Keller, 1993 ). In today’s phygital world ( Batat, 2022 ; Mele and Russo-Spena, 2022 ), such representations are typically nurtured by both physical (e.g. experiences and word of mouth) and digital (e.g. social media and websites) channels.

Previous research has often used free association tasks to explore social constructs. Originating in psychology, this approach is based on the understanding that knowledge is stored in associative networks ( Anderson and Bower, 1974 ). Associations are seen as valid indicators of meaning and were already used by Freud ( Rozin et al. , 2012 ). Association techniques can be applied in any research on social constructs, e.g. to investigate stereotyped representations of a family business ( Jaufenthaler, 2023 ), or consumers’ gender-related representations of meat ( Rozin et al. , 2012 ).

To analyze the resulting unstructured data, researchers have used various approaches ranging from simple hand counting to more sophisticated approaches, such as brand concept maps ( John et al. , 2006 ) or Vergès matrices ( Vergès, 1992 ). However, these traditional methods have notable limitations, including being time-consuming and providing limited insights, challenges that are more pronounced when dealing with large-scale data sets of complex nature. In this third application area, we argue that SemNA offers a more comprehensive solution by providing snapshots of today’s digitally nurtured social constructs in a profound, yet intuitive manner, graphically enabling deeper interpretations by revealing the centrality of nodes, semantic relationships and thematic clusters.

As a use case, we chose to explore consumers’ mental representation of Cristiano Ronaldo – one of the world’s most valuable human brands ( Forbes, 2022 ). Ronaldo is a professional football player renowned for his endorsements with global brands, such as Nike, and for winning several prestigious trophies at both the team and individual levels. In 2022, he became the first person to surpass the 400 million followers milestone on Instagram. Therefore, Ronaldo represents an interesting use case to illustrate a phygital social construct where mental associations are an amalgamation of his physical accomplishments on the field and his meticulously crafted social media persona.

We conducted a survey among Austrian students in December 2022. Participants were asked to list three to seven associations that came to mind upon hearing “Cristiano Ronaldo,” followed by providing their demographic information. A total of 824 participants completed the survey ( M age = 24 years; 30% female), and provided 3,180 associations. We engaged in basic preprocessing of the data set. The preprocessing involved cleaning the data set of typos and identifying semantic concepts, which are interrelated terms forming a single concept for analysis, such as “world footballer.” Consequently, we removed spaces between such words. We then used Wordij 3.0 ( Danowski, 2013a ; Wordij, 2024 ), an open-source software, for ATA to quantify the frequency of each association and to identify co-occurrences between association pairs. Co-occurrences were only counted within the different associations of a single respondent, necessitating the placement of a period at the end of each association set and setting the window size for the analysis at the sentence level. The applied settings also included a stop word list ( Humphreys and Wang, 2018 ), which eliminated words such as “and,” “because” and “example” from the data set. Additionally, numbers and punctuation within words were removed, words and word pairs that occurred fewer than two times were excluded, word pair order was not preserved and English contracted forms were expanded. This procedure resulted in 190 unique words (nodes) and 262 connections (edges) between the words. We then used Gephi ( Bastian et al. , 2009 ) to calculate the degree, and run the modularity algorithm (resolution: 1.0; modularity score: 0.174) for community detection ( Brandes et al. , 2008 ). We scaled the nodes and edges based on degree and weight, i.e. the larger the degree, the larger the node; the greater the weight, the thicker the edge. To improve readability, we filtered out all nodes with a degree below 3, resulting in visibility for 20.53% of nodes and 80.35% of edges ( Figure 4 ).

SemNA identified four distinct clusters, each encapsulating different facets of Cristiano Ronaldo’s public persona. The largest cluster (blue, 15%) includes associations related to Ronaldo’s former and current teams, as well as his main competitions. Core associations within this cluster are “realmadrid” (29), “portugal” (40) and “manchesterunited” (32). Lionel Messi (“messi” [24]) appears as another core association with links to associations such as “ballondor” (14) or “championsleague” (11). The green cluster (12%) centers on associations describing who Ronaldo is, featuring a blend of his professional accomplishments and personal life. The core association within this cluster is “worldfootballer” (32), representing a significant individual achievement for football players. Other associations focus on fame (e.g. “famous” [6], “rolemodel” [3], “star” [6]) and success (e.g. “winner” [4], “money” [9], “success” [8]), as well as associations regarding his family, including his wife (“georgina” [9]) and “children” (4). In the pink cluster (11%), key attributes and personality traits that distinguish Ronaldo are highlighted, including “hardwork” (8), “discipline” (8), “goals” (16), which surround the two most central associations “goat” (26) and “siuuu” (32). Interestingly, “siuuu,” referring to his iconic cheer, is linked to both positive (“worldfootballer” [32]) and negative associations (“arrogant” [31]) – the latter making up the red cluster (5%).

These insights enable a deeper interpretation of Ronaldo’s social construct, encompassing its phygital composition. To elaborate, prominent clusters primarily consist of co-occurring associations related to his former clubs (“realmadrid”), main competitions (“championsleague”) and his enduring rivalry with Messi (“messi”). Especially the latter, forming a cross-cluster relationship to “goat,” illustrates the phygital nature of Ronaldo’s social construct. That is, the contentious debate on who is the greatest of all time particularly unfolds on social media and online forums. The other clusters further demonstrate the role of his carefully curated social media presence in the formation of core associations with his persona. Specifically, the widespread association with “siuuu” stems from both his iconic celebration on the pitch and the viral memes and reels circulating on platforms like Instagram and TikTok ( Kassing, 2020 ). SemNA reveals that this association not only relates to positive but also to negative prejudice (“arrogance”). Other central aspects of the semantic network likewise highlight the phygital nature of Ronaldo’s personal brand. For example, the centrality of family-related associations reflects Ronaldo’s and his wife’s (“georgina”) frequent social media posts, which offer glimpses into their family life ( Jorge et al. , 2022 ), solidifying his association with being a family man.

Discussion application area 3.

This use case highlights the versatility of SemNA and its potential in examining phygital social constructs, seamlessly digitizing cognitive associations to form a holistic and tangible digital representation. Consequently, SemNA’s output can be used for in-depth explorations and interpretations of facets of social constructs, including the intertwined physical and digital compositions inherent in today’s phygital world. For example, the intuitive overview of associations’ centrality through node size reveals the prominence of Ronaldo’s iconic cheer “siuuu” within the investigated audience, shaped through both physical and digital channels ( Kassing, 2020 ). Furthermore, insights into cross-cluster relationships, such as between “goat” and “messi,” uncover underlying connections that may otherwise remain hidden but are crucial for forming a deeper understanding of phygital social constructs. Overall, this use case exemplifies a new application area for SemNA in consumer and marketing research, offering valuable insights by capturing comprehensive snapshots of complex social constructs.

Consumer and marketing research increasingly demands sophisticated methodologies capable of analyzing large-scale text-based data ( Humphreys and Wang, 2018 ). In particular, the confluence of physical and digital environments ( Batat, 2023 ) poses methodological challenges for researchers addressing the hybrid nature of experiences and consumption phenomena at this intersection. Contributing to these timely discussions, we demonstrate that SemNA represents a well-suited method to address challenges in this evolving landscape as it supports researchers in exploring, understanding and accounting for the nature, size, scope and relationships between physical and digital elements of large-scale text-based data.

This paper illustrates the versatility of SemNA as a methodology that can be applied to various research phenomena and text-based data sources in phygital contexts ( Batat, 2022 , 2023 ) by presenting three application areas and use cases for SemNA in consumer and marketing research. The first application area uses SemNA to investigate the complex relationships between digital and physical consumption practices, as exemplified by analyzing narrative interviews about fashion content consumption on Instagram and its influence on physical consumption. The second application area uses SemNA to explore the phygital nature of social media discourse, trends and practices through consumer-generated content. The third application area uses SemNA to convert phygital social constructs into tangible digital representations for subsequent in-depth exploration, as illustrated by the use case of Cristiano Ronaldo. This research contributes to recent fundamental work on computational text analysis methods in consumer and marketing research ( Berger et al. , 2020 , 2022 ; Humphreys and Wang, 2018 ) by focusing on SemNA as a computational method that centers on the visual exploration and interpretation of relationships and representations of knowledge, information, concepts and meanings in large-scale text-based data. Drawing on prior work from related disciplines that delineate general procedural steps for SemNA ( Christensen and Kenett, 2021 ; Segev, 2021 ), our work offers guidance on when , where and how SemNA holds valuable potential for consumer and marketing research unfolding in contemporary phygital contexts. In doing so, we provide important steps toward facilitating the adoption, application and impact of SemNA in consumer and marketing research by outlining detailed accounts of application areas and demonstrating usage and interpretive analyses through specific use cases.

By highlighting this qualitative, interpretivist approach to SemNA, this research also contributes to pioneering discussions on integrating tool-based methods into qualitative, interpretivist methodologies ( Caliandro and Gandini, 2017 ). Accordingly, this research opens methodological pathways for qualitative researchers that may encounter the limitations of both traditional and specialized qualitative research methods when analyzing large-scale text-based data in various research contexts. The outlined approach mitigates these limitations by using a quant-qual approach to SemNA, in which the quantitative part structures and visualizes the data, whereas the subsequent qualitative part then engages in data-driven, interpretivist exploration and theorization ( Lucarelli et al. , 2023 ). As such, this approach provides qualitative, interpretivist researchers with the methodological means to properly grasp the nature, size and scope of unstructured, large-scale text-based data while maintaining an overarching interpretivist spirit.

Limitations and future research

Understanding the limitations of this research opens up opportunities for future research. First, this research exclusively focuses on the interactions between the physical world and social media. Future research should consider the broader spectrum of digital interactions outside social media platforms. Second, while this study proposes three broad application areas for SemNA, this selection may not fully capture the method’s versatility and potential applications across different domains or contexts. We encourage future research to explore further application areas, supporting researchers in uncovering diverse insights and demonstrating the method’s wider applicability and effectiveness in analyzing complex text-based data in phygital contexts. Third, the primary aim of the paper is to illustrate when , where and how SemNA can be used, therefore, prioritizing methodological exposition over comprehensive, context-rich interpretation. Researchers conducting empirically-driven studies to systematically investigate a particular phenomenon should engage in deeper interpretations. Fourth, this study uses only basic data cleaning procedures, such as removing stop words and does not use more nuanced preprocessing techniques, such as lemmatization or stemming ( Balakrishnan and Lloyd-Yemoh, 2014 ), which may hinder the accurate representation and analysis of linguistic relationships in certain contexts. Finally, the qualitative, interpretivist approach to SemNA, while promising, is relatively new. More empirical research and methodological experimentation will be beneficial in identifying the optimal applications of this approach in various research contexts. This highlights the need for additional studies to refine the methodological framework, improve analytical techniques and establish best practices for using SemNA in consumer and marketing research. General limitations of SemNA as a method must also be considered, including its inability to substitute for thorough reading and textual understanding, its challenge in fully capturing the context of word relationships without additional contextual intelligence by researchers and its incapacity to interpret linguistic ambiguities independently.

This paper aims to provide a foundation for the adoption and utilization of SemNA in consumer and marketing research. To facilitate this adoption, we illustrated possible research inquiries, including exemplary research questions and data sources, for each application area in Figure 1 . Future research could use SemNA, for example, to investigate experiences and consumption phenomena in other phygital contexts, such as how the integration of augmented reality in retail environments influences consumer decision-making processes; to investigate how consumers narrate their physical consumption practices related to a specific discourse or lifestyle, such as sustainable fashion, on digital platforms; or to capture phygital social constructs across diverse segments with distinct digital behaviors. In general, we invite future research to join our endeavor by identifying and elaborating on further application areas and use cases for SemNA in consumer and marketing research, particularly in combination with artificial intelligence to potentially enhance SemNA’s capabilities and automate aspects of the data analysis process.

The application areas outlined for SemNA demonstrate its utility for marketing and brand managers in gaining a comprehensive understanding of complex, multidimensional ecosystems composed of both physical and digital elements, including various relevant channels and stakeholders. SemNA empowers managers to capture and integrate company- and brand-related associations within their communication strategies, effectively addressing consumer-initiated trends and practices. Correspondingly, managers can use SemNA to acquire cross-channel and platform intelligence. Using SemNA fosters a consumer-centric, co-creative approach, positioning consumers as active participants in shaping the company’s or brand’s image.

Using SemNA to extract insights from text-based consumer data enables managers to “design appropriate and satisfying physical experiences while responding to consumers’ functional, emotional, social, sensory, and symbolic needs across different consumption fields” ( Batat, 2022 : 8). For example, consumers often communicate their experiences via social media ( Klostermann et al. , 2018 ). Managers can leverage SemNA for real-time analysis and visualization of attitudes, associations or sentiments expressed on social media, allowing for prompt and informed decision-making. Additionally, managers can digitize data gathered in the physical world and translate them into the channel/platform language for their communication campaigns, e.g. hashtagging core associations on social media, thereby positioning the brand/company in the corresponding network. Similarly, SemNA enables managers to gain an increased understanding of brand perceptions (and other social constructs) and strategically manage brand reputation by analyzing corresponding associations and meaning clusters ( Jaufenthaler et al. , 2023 ).

Application areas for semantic network analysis in consumer and marketing research

Semantic network analysis of interview data

Semantic network of hashtags co-occurring with #sustainability on Instagram

Capturing the phygital social construct ‘Cristiano Ronaldo’

Element (synonyms)	Description	Measure unit	Network visualization representation
Node (actors and vertices)	that represents a unique element in the semantic network ( )	Describes the centrality/significance of the node within a network and is measured by counting the total number of edges connected to a node	Its size depends on the calculated degree of the respective node
Edge (links, ties and connections)	Describes the of two or more semantic units/concepts, indicating the relationship, similarity or strength of the associations between two or more nodes ( )	Represents the strength of the relationship between connected nodes and is typically determined by the frequency of the co-occurrence	Its thickness depends on the weight of the edge
Modularity classes (clusters and communities)	The outcome of an algorithmic process that compartmentalizes a network into thematic based on densely connected nodes within the network ( )	Describes the size of a respective modularity class relative to the entire network	Each is uniformly colored

Source: Author's own work

Anderson , J.R. and Bower , G.H. ( 1974 ), “ A propositional theory of recognition memory ”, Memory and Cognition , Vol. 2 No. 3 , pp. 406 - 412 .

Balakrishnan , V. and Lloyd-Yemoh , E. ( 2014 ), “ Stemming and lemmatization: a comparison of retrieval performances ”, Lecture Notes on Software Engineering , Vol. 2 No. 3 , pp. 262 - 267 .

Bastian , M. , Heymann , S. and Jacomy , M. ( 2009 ), “ Gephi: an open source software for exploring and manipulating networks ”, Proceedings of the International AAAI Conference on Web and Social Media , Vol. 3 No. 1 , pp. 361 - 362 .

Batat , W. ( 2022 ), “ What does phygital really mean? A conceptual introduction to the phygital customer experience (PH-CX) framework ”, Journal of Strategic Marketing , pp. 1 - 24 .

Batat , W. ( 2023 ), “ Experiential research as a methodological framework for studying consumer behaviors in phygital settings ”, Qualitative Market Research: An International Journal , Vol. 26 No. 3 , pp. 269 - 277 .

Beichert , M. , Bayerl , A. , Goldenberg , J. and Lanz , A. ( 2024 ), “ Revenue generation through influencer marketing ”, Journal of Marketing , p. 00222429231217471 .

Berger , J. , Humphreys , A. , Ludwig , S. , Moe , W.W. , Netzer , O. and Schweidel , D.A. ( 2020 ), “ Uniting the tribes: using text for marketing insight ”, Journal of Marketing , Vol. 84 No. 1 , pp. 1 - 25 .

Berger , J. , Packard , G. , Boghrati , R. , Hsu , M. , Humphreys , A. , Luangrath , A. , Moore , A. , Nave , G. , Olivola , C. and Rocklage , M. ( 2022 ), “ Marketing insights from text analysis ”, Marketing Letters , Vol. 33 No. 3 , pp. 365 - 377 .

Blondel , V.D. , Guillaume , J.-L. , Lambiotte , R. and Lefebvre , E. ( 2008 ), “ Fast unfolding of communities in large networks ”, Journal of Statistical Mechanics: Theory and Experiment , Vol. 2008 No. 10 , p. P10008 .

Brandes , U. , Delling , D. , Gaertler , M. , Görke , R. , Hoefer , M. , Nikoloski , Z. and Wagner , D. ( 2008 ), “ On modularity clustering ”, IEEE Transactions on Knowledge and Data Engineering , Vol. 20 No. 2 , pp. 172 - 188 .

Bruns , A. and Snee , H. (Eds) ( 2022 ), “ How to visually analyse networks using gephi [how-to guide] ”, Sage Research Methods: Doing Research Online , Sage , Los Angeles, CA .

Caliandro , A. and Anselmi , G. ( 2021 ), “ Affordances-based brand relations: an inquire on memetic brands on Instagram ”, Social Media + Society , Vol. 7 No. 2 , p. 20563051211021367 .

Caliandro , A. and Gandini , A. ( 2017 ), Qualitative Research in Digital Environments: A Research Toolkit , Routledge , London .

Christensen , A.P. and Kenett , Y.N. ( 2021 ), “ Semantic network analysis (SemNA): a tutorial on preprocessing, estimating, and analyzing semantic networks ”, Psychological Methods , Vol. 28 No. 4 , pp. 860 - 879 .

Danowski , J.A. ( 1988 ), “ Organizational infographics and automated auditing: using computers to unobtrusively gather and analyze communication ”, in Goldhaber , G. and Barnett , G. (Eds), Handbook of Organizational Communication , Ablex , Norwood, NJ , pp. 385 - 434 .

Danowski , J.A. ( 2013a ), WORDij Version 3.0: Semantic Network Analysis Software , University of IL at Chicago , Chicago, IL .

Danowski , J.A. ( 2013b ), “ Social media network size and semantic networks for collaboration in design ”, International Journal of Organisational Design and Engineering , Vol. 2 No. 4 , pp. 343 - 361 .

Doerfel , M.L. ( 1998 ), “ What constitutes semantic network analysis? A comparison of research and methodologies ”, Connections , Vol. 21 No. 2 , pp. 16 - 26 .

Drieger , P. ( 2013 ), “ Semantic network analysis as a method for visual text analytics ”, Procedia – Social and Behavioral Sciences , Vol. 79 , pp. 4 - 17 .

Forbes ( 2022 ), “ Cristiano Ronaldo ”, available at: www.forbes.com/profile/cristiano-ronaldo/?sh=34f3b2b7565d ( accessed 27 February 2023 ).

Gephi ( 2024 ), “ Gephi ”, available at: https://gephi.org/

Gibbs , M. , Meese , J. , Arnold , M. , Nansen , B. and Carter , M. ( 2015 ), “ #funeral and Instagram: death, social media, and platform vernacular ”, Information, Communication and Society , Vol. 18 No. 3 , pp. 255 - 268 .

Heisley , D.D. and Levy , S.J. ( 1991 ), “ Autodriving: a photoelicitation technique ”, Journal of Consumer Research , Vol. 18 No. 3 , pp. 257 - 272 .

Humphreys , A. and Wang , R.J.H. ( 2018 ), “ Automated text analysis for consumer research ”, Journal of Consumer Research , Vol. 44 No. 6 , pp. 1274 - 1306 .

Jaufenthaler , P. ( 2023 ), “ Beauty lies in the eye of the beholder: consumers’ and jobseekers’ interpretations of the family business brand ”, Journal of Product and Brand Management , Vol. 32 No. 5 , pp. 697 - 712 .

Jaufenthaler , P. , Koll , O. , Lude , M. and Prügl , R. ( 2023 ), “ Country differences in family firm reputation: an exploration in Germany, India, and the United States ”, Family Business Review , Vol. 36 No. 4 , pp. 352 - 374 .

John , D.R. , Loken , B. , Kim , K. and Monga , A.B. ( 2006 ), “ Brand concept maps: a methodology for identifying brand association networks ”, Journal of Marketing Research , Vol. 43 No. 4 , pp. 549 - 563 .

Jorge , A. , Marôpo , L. and Neto , F. ( 2022 ), “ When you realise your dad is cristiano ronaldo’: celebrity sharenting and children’s digital identities ”, Information, Communication and Society , Vol. 25 No. 4 , pp. 516 - 535 .

Kassing , J.W. ( 2020 ), “ Messi hanging laundry at the Bernabéu: the production and consumption of internet sports memes as trash talk ”, Discourse, Context and Media , Vol. 34

Keller , K.L. ( 1993 ), “ Conceptualizing, measuring, and managing customer-based brand equity ”, Journal of Marketing , Vol. 57 No. 1 , pp. 1 - 22 .

Klostermann , J. , Plumeyer , A. , Böger , D. and Decker , R. ( 2018 ), “ Extracting brand information from social networks: integrating image, text, and social tagging data ”, International Journal of Research in Marketing , Vol. 35 No. 4 , pp. 538 - 556 .

Ko , E. , Chun , E. , Song , S. and Mattila , P. ( 2015 ), “ Exploring SNS as a consumer tool for retail therapy: explicating semantic networks of ‘shopping makes me happy (unhappy)’ as a new product development method ”, Journal of Global Scholars of Marketing Science , Vol. 25 No. 1 , pp. 37 - 48 .

Kozinets , R. ( 2020 ), Netnography: The Essential Guide to Qualitative Social Media Research , Sage , London .

Lehmann , F. ( 1992 ), “ Semantic networks ”, Computers and Mathematics with Applications , Vol. 23 Nos 2/5 , pp. 1 - 50 .

Lucarelli , A. , Shahriar , H. , Ulver , S. and Egan-Wyer , C. ( 2023 ), “ Research contributions in interpretivist marketing and consumer research studies: a kaleidoscopic framework ”, Marketing Theory , p. 14705931231202430 .

Marres , N. ( 2015 ), “ Why map issues? On controversy analysis as a digital method ”, Science, Technology, and Human Values , Vol. 40 No. 5 , pp. 655 - 686 .

Mele , C. and Russo-Spena , T. ( 2022 ), “ The architecture of the phygital customer journey: a dynamic interplay between systems of insights and systems of engagement ”, European Journal of Marketing , Vol. 56 No. 1 , pp. 72 - 91 .

Merriam-Webster Dictionary ( 2024 ), “ Social construct ”, available at: www.merriam-webster.com/dictionary/social%20construct ( accessed 28 February 2024 ).

Netzer , O. , Feldman , R. , Goldenberg , J. and Fresko , M. ( 2012 ), “ Mine your own business: market-Structure surveillance through text mining ”, Marketing Science , Vol. 31 No. 3 , pp. 521 - 543 .

Packard , G. and Berger , J. ( 2023 ), “ The emergence and evolution of consumer language research ”, Journal of Consumer Research , p. ucad013 .

Rice , R.E. ( 2005 ), “ New media/internet research topics of the association of internet researchers ”, The Information Society , Vol. 21 No. 4 , pp. 285 - 299 .

Rozin , P. , Hormes , J.M. , Faith , M.S. and Wansink , B. ( 2012 ), “ Is meat male? A quantitative multimethod framework to establish metaphoric relationships ”, Journal of Consumer Research , Vol. 39 No. 3 , pp. 629 - 643 .

Schöps , J.D. , Kogler , S. and Hemetsberger , A. ( 2020 ), “ (De-)stabilizing the digitized fashion market on Instagram–dynamics of visual performative assemblages ”, Consumption Markets and Culture , Vol. 23 No. 2 , pp. 195 - 213 .

Schöps , J.D. , Reinhardt , C. and Hemetsberger , A. ( 2022 ), “ Sticky market webs of connection – human and nonhuman market co-codification dynamics across social media ”, European Journal of Marketing , Vol. 56 No. 13 , pp. 78 - 104 .

Schöps , J.D. , Schwarz , S. and Rojkowski , V. ( 2024 ), “ Memetic logics of participation: fitness body culture on Instagram ”, Marketing Theory , Vol. 24 No. 1 , pp. 65 - 89 .

Schröder , J. ( 2018 ), “ Build your own Instagram database ”, available at: https://towardsdatascience.com/build-your-own-instagram-database-134281e8ee92

Segev , E. ( 2021 ), Semantic Network Analysis in Social Sciences , Routledge , London .

Teichert , T.A. and Schöntag , K. ( 2010 ), “ Exploring consumer knowledge structures using associative network analysis ”, Psychology and Marketing , Vol. 27 No. 4 , pp. 369 - 398 .

Vergès , P. ( 1992 ), “ L’évocation de l’argent: une méthode pour la définition du noyau Central d'une représentation ”, Bulletin De Psychologie , Vol. 45 No. 405 , pp. 203 - 209 .

Wordij ( 2024 ), “ Wordij ”, available at: https://wordij.net/

Acknowledgements

All authors contributed equally to this paper and should be regarded as joint first authors.

Co-first authors can prioritize their names when adding this paper’s reference to their resumes.

Corresponding author

About the authors.

Jonathan David Schöps is an Assistant Professor of Marketing at Stockholm Business School, Stockholm University. His research focuses on digital market dynamics and the role of digital technologies and nonhuman actors in market shaping. Methodologically, Jonathan specializes not only in digital methods, such as semantic network analysis or automated text analysis and visual methodologies but also holds expertise in traditional qualitative research methods.

Philipp Jaufenthaler is a postdoctoral marketing researcher at the University of Innsbruck, specializing in brand localness and the reputation of family firms. His research uses a versatile methodological approach, spanning from quantitative experimental studies to qualitative and mixed-method analyses. Beyond academia, he contributes his expertise as a strategy consultant, enhancing the practical applications of his research findings in the business world.

All feedback is valuable.

Please share your general feedback

Report an issue or find answers to frequently asked questions

Contact Customer Support

Home » Social Network Analysis – Types, Tools and Examples

Social Network Analysis – Types, Tools and Examples

Table of Contents

Social Network Analysis

Social Network Analysis (SNA) is an analytical method used to study social structures through the use of networks and graph theory. It identifies the relationships between individuals, organizations, or other entities and examines the patterns and implications of these relationships.

The nodes in the network represent the actors within the networks and the ties or edges represent relationships between the actors. These might be, for example, friendship ties between people, business relationships between companies, or communication patterns between individuals.

By analyzing the network structure and the characteristics of the actors within the network, SNA can reveal properties such as the distribution of resources, the flow of information, or the overall connectivity of the network.

Here are a few key concepts in SNA:

Centrality : This measures the importance of a node in the network. Various centrality measures exist, each emphasizing a different aspect of a node’s position within the network, such as degree centrality (the number of direct connections a node has), betweenness centrality (the number of times a node acts as a bridge along the shortest path between two other nodes), and eigenvector centrality (the sum of the centrality scores of all nodes that one node is connected to).
Density : This is a measure of the proportion of possible connections in a network that are actual connections. A high density suggests that the network participants are highly interconnected.
Clusters or Communities : These are groups of nodes that are more densely connected with each other than with the rest of the network.
Structural Holes : These are gaps in the network where a node could potentially act as a bridge between two unconnected parts of the network.

Types of Social Network Analysis

Social Network Analysis can be broadly categorized based on the type of networks being analyzed, the level of analysis, and the methodologies employed. Here are a few ways to categorize SNA:

Whole Network Analysis

This type of analysis focuses on the structure and properties of the network as a whole. This might include measures of network cohesion, centralization, and density. It also looks at the overall distribution of relationships and identifies key groups or clusters within the network.

Ego Network Analysis

In this type of analysis, the focus is on a single actor (the ‘ego’) and their immediate network (the ‘alters’). It’s often used when interest is in the personal networks of individuals. Measures can include the size of the network, the composition of the network in terms of the types of ties and nodes, and measures of network density or diversity.

Two-mode (or Bipartite) Network Analysis

This type of SNA is used when there are two different types of nodes, and connections are only possible between nodes of different types (not within types). For example, authors and the books they write, or actors and the movies they appear in. In such a network, you can study the connections between nodes of one type, mediated by nodes of the other type.

Dynamic Network Analysis (DNA)

This is used to study how social networks evolve over time. This could involve studying how ties between actors develop or disappear, or how actors move around within the network. In addition to traditional network measures, DNA also considers measures that are dynamic in nature, such as change in centrality over time.

Semantic Network Analysis

This type of SNA focuses on the relationships between concepts or ideas, rather than individuals or organizations. For instance, semantic network analysis could map out how different scientific concepts are related to each other in the literature.

Social Media Network Analysis

A specialized form of SNA, this deals with the study of social relationships as expressed through social media platforms. It allows for the mapping and measuring of relationships and flows between people, groups, organizations, computers, URLs, and other connected information/knowledge entities.

Social Network Analysis Techniques

Social Network Analysis involves various techniques to understand the structure and patterns of relationships among actors (people, organizations, etc.) in a network. These techniques may be mathematical, visual, or computational, and often involve the use of specialized software. Here are several common SNA techniques:

Network Visualization

One of the most basic SNA techniques involves creating a visual representation of the network. This can help to reveal patterns and structures within the network that may not be immediately obvious from the raw data. There are various ways to create such visualizations, depending on the specifics of the network and the goals of the analysis. Software such as Gephi or Cytoscape can be used for network visualization.

Centrality Measures

These are techniques used to identify the most important nodes within a network. Various measures of centrality exist, each highlighting different aspects of a node’s position in the network. These include degree centrality (the number of connections a node has), betweenness centrality (how often a node appears on the shortest path between other nodes), closeness centrality (how quickly a node can reach all other nodes in the network), and eigenvector centrality (a measure of the influence of a node in a network).

Community Detection

Also known as clustering, this technique aims to identify groups of nodes that are more closely connected with each other than with the rest of the network. This can help to reveal sub-groups or communities within the network.

Structural Equivalence and Blockmodeling

Structural equivalence is a measure of how similarly two nodes are connected to the rest of the network. Nodes that are structurally equivalent often play similar roles in the network. Blockmodeling is a technique used to simplify a network by grouping together structurally equivalent nodes.

Dynamic Network Analysis

This involves studying how a network changes over time. This can help to reveal patterns of network evolution, including how relationships form and dissolve, how centrality measures change over time, and how communities evolve.

Network Correlation and Regression

These are statistical techniques used to identify and test for patterns within the network. For example, one might use these techniques to test whether nodes with certain characteristics are more likely to form connections with each other.

Social Network Analysis Tools

There are several tools available that can be used to conduct Social Network Analysis (SNA). These range from open-source software to commercial offerings, each with their own strengths and weaknesses. Here are a few examples:

Gephi : Gephi is an open-source, interactive visualization and exploration platform for all kinds of networks and complex systems. It’s user-friendly and allows users to interactively manipulate the network visualization, perform network analysis, and export results in various formats.
UCINet : UCINet is a comprehensive package for the analysis of social network data as well as other 1-mode and 2-mode data. It’s widely used in social science research.
NetDraw : Often used in conjunction with UCINet, NetDraw is a free tool for visualizing networks. It supports the visualization of large networks and allows for various customization options.
Pajek : Pajek is a program for the analysis and visualization of large networks. It’s an extensive tool, offering a range of complex network metrics, and is free for non-commercial use.
NodeXL : NodeXL is a free, open-source template for Microsoft Excel that allows users to display and analyze network graphs. Its integration with Excel makes it user-friendly, particularly for those already familiar with Excel.
Cytoscape : Originally designed for biological research, Cytoscape is now a popular open-source software platform for visualizing complex networks and integrating these with any type of attribute data.
SocioViz : SocioViz is a social media analytics platform for Twitter data, focused on network analysis and visualization. It’s a powerful tool for researchers interested in online social networks.
NetworkX : NetworkX is a Python package for the creation, manipulation, and study of the structure, dynamics, and functions of complex networks. It integrates well with other scientific Python tools like SciPy and Matplotlib.
igraph : igraph is a library available in R, Python, and C for creating, manipulating, and analyzing networks. It’s highly efficient and can handle large networks.
RSiena : RSiena is an R package dedicated to the statistical analysis of network data, with a particular focus on longitudinal social networks.

Social Network Analysis Examples

Social Network Analysis Examples are as follows:

Public Health – COVID-19 Pandemic : During the COVID-19 pandemic, SNA was used to model the spread of the virus. The interactions between individuals were mapped as a network, helping identify super-spreader events and informing public health interventions.
Business – Google’s “PageRank” Algorithm : Google’s PageRank algorithm, which determines the order of search engine results, is a type of SNA. It considers web pages as nodes and hyperlinks as connections, determining a page’s importance by looking at the number and quality of links to it.
Sociology – Stanley Milgram’s “Small World” Experiment : This is one of the most famous social network experiments, where Milgram demonstrated that any two people in the United States are separated on average by only six acquaintances, leading to the phrase “six degrees of separation.”
Online Social Networks – Facebook’s “People You May Know” Feature : Facebook uses SNA to suggest new friends. The platform analyzes your current network and suggests people you’re likely to know, typically friends of friends or people who share common networks.
Criminal Network Analysis – Capture of Osama bin Laden : SNA was reportedly used in the operation to capture Osama bin Laden. By mapping the social connections of known associates, intelligence agencies were able to locate the Al-Qaeda leader.
Academic Research – Collaboration Networks : SNA is used in scientometrics to analyze collaboration networks among researchers . For example, a study on co-authorship networks in scientific articles can reveal patterns of collaboration and the flow of information in different disciplines.

When to use Social Network Analysis

Social Network Analysis is a powerful tool for studying the relationships between entities (like people, organizations, or even concepts) and the overall structure of these relationships. Here are several situations when SNA might be particularly useful:

Understanding Complex Systems : SNA is well-suited to studying complex, interconnected systems. If you’re interested in not just individual entities but also the relationships between them, SNA can provide valuable insights.
Identifying Key Actors : SNA can help identify the most important entities in a network based on their position and connections. These might be influential people within a social network, critical servers in a computer network, or key scholars in a field of study.
Studying Diffusion Processes : If you’re interested in how something (like information, behaviors, diseases) spreads through a network, SNA can be a valuable tool. It allows for the examination of diffusion pathways and identification of nodes that speed up or hinder diffusion.
Detecting Communities : SNA can be used to identify clusters or communities within a network. These might be groups of friends within a social network, clusters of companies in a business network, or research clusters in scientific collaboration networks.
Mapping Out Large Systems : In cases where you have a large system of many interconnected entities, SNA can provide a visual representation of the system, making it easier to understand and analyze.
Investigating Structural Roles : If you’re interested in the roles individuals or entities play within their network, SNA offers methods to classify these roles based on the pattern of their relationships.

Purpose of Social Network Analysis

Social Network Analysis serves a wide range of purposes across different fields, given its versatile nature. Here are several key purposes:

Understanding Network Structure : One of the key purposes of SNA is to understand the structure of relationships between actors within a network. This includes understanding how the network is organized, the distribution of connections, and the patterns of interaction.
Identifying Key Actors or Nodes : SNA can identify crucial nodes within a network. These could be individuals with many connections, or nodes that serve as critical links between different parts of the network. In a business, for instance, such nodes might be key influencers or innovators.
Detecting Subgroups or Communities : SNA can identify clusters or communities within a network, i.e., groups of nodes that are more connected to each other than to the rest of the network. This can be valuable in numerous contexts, from identifying communities in social media networks to detecting collaboration clusters in scientific networks.
Analyzing Information or Disease Spread : In public health and communication studies, SNA is used to study the patterns and pathways of information or disease spread. Understanding these patterns can be critical for designing effective interventions or campaigns.
Analyzing Social Capital : SNA can help understand an individual or group’s social capital – the resources they can access through their network relationships. This analysis can offer insights into power dynamics, access to resources, and inequality within a network.
Studying Network Dynamics : SNA can examine how networks change over time. This could involve studying how relationships form or dissolve, how centrality measures change over time, or how communities evolve.
Predicting Future Interactions : SNA can be used to predict future interactions or relationships within a network, which can be useful in a variety of settings such as recommender systems, predicting disease spread, or forecasting emerging trends in social media.

Applications of Social Network Analysis

Social Network Analysis has a wide range of applications across different disciplines due to its capacity to analyze relationships and interactions. Here are some common areas where it is applied:

Public Health : SNA can be used to understand the spread of infectious diseases within a community or globally. It helps identify “super spreaders” and optimizes strategies for vaccination or containment.
Business and Organizations : Companies use SNA to analyze communication and workflow patterns, enhance collaboration, boost efficiency, and detect key influencers within their organization. It can also be applied in understanding and leveraging informal networks within a business.
Social Media Analysis : On platforms like Facebook, Twitter, or Instagram, SNA helps analyze user behavior, track information dissemination, identify influencers, detect communities, and develop recommendation systems.
Criminal Justice : Law enforcement and intelligence agencies use SNA to understand the structure of criminal or terrorist networks, identify key figures, and predict future activities.
Internet Infrastructure : SNA helps in mapping the internet, identifying critical nodes, and developing strategies for robustness against cyberattacks or outages.
Marketing : In marketing, SNA can track the diffusion of advertising messages, identify influential consumers for targeted marketing, and understand consumer behavior and brand communities.
Scientometrics : SNA is used in academic research to map co-authorship networks or citation networks. It can uncover patterns of collaboration and the flow of knowledge in scientific fields.
Politics and Policy Making : SNA can help understand political alliances, lobby networks, or policy networks, which can be critical for strategic decision-making in politics.
Ecology : In ecological studies, SNA can help understand the relationships between different species in an ecosystem, providing valuable insights into ecological dynamics.

Advantages of Social Network Analysis

Social Network Analysis offers several advantages when studying complex systems and relationships. Here are a few key advantages:

Reveals Complex Relationships : SNA allows for the study of relationships between entities (be they people, organizations, computers, etc.) in a way that many other methodologies do not. It emphasizes the importance of these relationships and helps reveal complex interaction patterns.
Identifies Key Players : SNA can identify the most influential or important nodes in a network, whether they are individuals within a social network, key servers in an internet network, or central scholars in an academic field.
Unveils Network Structure and Communities : SNA can help visualize the overall structure of a network and can reveal communities or clusters of nodes within a network. This can provide valuable insights into the organization and division of a network.
Tracks Changes Over Time : Dynamic SNA allows the study of networks over time. This can help to track changes in the network structure, the role of specific nodes, or the flow of information or resources through the network.
Helps Predict Future Interactions : Based on the analysis of current and past relationships, SNA can be used to predict future interactions, which can be useful in many fields including public health, marketing, and national security.
Aids in Designing Effective Strategies : The insights gained from SNA can be used to design targeted strategies, whether that’s intervening in the spread of misinformation online, designing a targeted marketing campaign, disrupting a criminal network, or managing collaboration in an organization.
Versatility : SNA can be applied to a vast array of fields, from sociology to computer science, biology to business, making it a versatile tool.

Disadvantages of Social Network Analysis

While Social Network Analysis is a powerful tool with wide-ranging applications, it also has certain limitations and disadvantages that are important to consider:

Data Collection Challenges : Collecting complete and accurate network data can be a major challenge. For larger networks, it may be nearly impossible to collect data on all relevant relationships. There’s also a risk of response bias, as people may forget, overlook, or misinterpret their relationships when providing data.
Time and Resource Intensive : Collecting network data, especially from primary sources, can be extremely time-consuming and expensive. Additionally, analyzing network data can also require significant computational resources for larger networks.
Complexity : SNA involves complex concepts and measures, which can be difficult to understand without specialized knowledge. This complexity can make it difficult to communicate findings to a non-technical audience.
Privacy and Ethical Concerns : SNA often involves sensitive data about individuals’ relationships and interactions, raising important privacy and ethical concerns. It’s important to handle this data carefully to respect individuals’ privacy.
Static Snapshots : Traditional SNA often provides a static snapshot of a network at a particular point in time, which may not capture the dynamic nature of social relationships. While dynamic SNA does exist, it adds additional complexity and data demands.
Dependence on Quality of Data : The insights and conclusions drawn from SNA are only as good as the data used. Incomplete, inaccurate, or biased data can lead to misleading results.
Difficulties in Establishing Causality : While SNA can reveal patterns and associations in network data, it can be difficult to establish causal relationships. For instance, do strong connections between two individuals lead to similar behavior, or does similar behavior lead to strong connections?
Assumptions about Relationships : SNA often assumes that relationships are equally important, which might not always be the case. Different relationships might have different strengths or meanings, which can be challenging to represent in a network.

About the author

Muhammad Hassan

Researcher, Academic Writer, Web developer

Predictive Analytics – Techniques, Tools and...

Sentiment Analysis – Tools, Techniques and...

Descriptive Analytics – Methods, Tools and...

What is Big Data? Types, Tools and Examples

Prescriptive Analytics – Techniques, Tools and...

Diagnostic Analytics – Methods, Tools and...

Analysis of Qualitative Data: Using Automated Semantic Analysis to Understand Networks of Concepts

First Online: 28 September 2017

Cite this chapter

Louise Young 3 , 4 &
Kristin B. Munksgaard 4

2057 Accesses

1 Citations

Qualitative research of business relationships and networks is of limited value if the analysis does not address the rich interdependencies of the processes and mechanisms involved and does not feature clear and credible methods of analysis. The purpose of this chapter is to consider approaches to analysis that provide opportunities for demonstrable, credible analysis of the qualitative nuances of business systems. The particular focus is on the use of computer-aided methods, including the Leximancer lexicographic analysis software, that, combined with more traditional methods, provide reliable and meaningful analysis of large quantities of textual information, including interview transcripts and secondary data. These processes and the findings that they can produce are demonstrated using a wide range of the authors’ own research.

This is a preview of subscription content, log in via an institution to check access.

Access this chapter

Subscribe and save.

Get 10 units per month
Download Article/Chapter or eBook
1 Unit = 1 Article or 1 Chapter
Cancel anytime
Available as PDF
Read on any device
Instant download
Own it forever
Available as EPUB and PDF
Compact, lightweight edition
Dispatched in 3 to 5 business days
Free shipping worldwide - see info
Durable hardcover edition

Tax calculation will be finalised at checkout

Purchases are for personal use only

Institutional subscriptions

Theory Building from the Quantitative Analysis of Qualitative Information

Abductive Thematic Network Analysis (ATNA) Using ATLAS-ti

Qualitative research: extending the range with flexible pattern matching

Ahuvia, A. 2001. Traditional, interpretive, and reception based content analyses: Improving the ability of content analysis to address issues of pragmatic and theoretical concern. Social Indicators Research 54 (2): 139–172.

Article Google Scholar

Angus-Leppan, Tamsin, Suzanne Benn, and Louise Young. 2010. A sensemaking approach to trade-offs and synergies between human and ecological elements of corporate sustainability. Journal of Business and Environment 19 (4): 230–244.

Google Scholar

Bairstow, Nigel, and Louise Young. 2012. How channels evolve, a historical explanation. Industrial Marketing Management 41 (1): 385–393.

Bonoma, Thomas V. 1985. Case research in marketing: Opportunities, problems, and a process. Journal of Marketing Research (JMR) 22 (2): 199–208.

Braun, Virginia, and Victoria Clark. 2006. Using thematic analysis in psychology. Qualitative Research in Psychology 3 (2): 77–101.

Burton, C., J. Louviere, and Louise Young. 2009. Retaining the visitor, enhancing the experience: Identifying attributes of choice in repeat museum visitation. Journal of Voluntary and Non Profit Marketing 14: 21–34.

Cova, B., and S. Pace. 2006. Brand community of convenience products: New forms of customer empowerment-the case “My Nutella, The Community”. European Journal of Marketing 40 (9/10): 1087–1105.

Daly, J., K. Willis, R. Small, J. Green, N. Welch, M. Kealy, and E. Hughes. 2007. A hierarchy of evidence for assessing qualitative health research. Journal of Clinical Epidemiology 60 (1): 43–49.

Danneels, E., and G.L. Lilien. 1998. Doctoral programs in business-to-business marketing: Status and prospects. Journal of Business-to-Business Marketing 5 (1–2): 7–34.

Dawson, Bonnie, John Murray, Ian Wilkinson, and Louise Young. 2009. Towards a more substantial examination of supplier-customer relationship performance. In Proceedings, IMP Conference, Marseilles, France , September 2009. www.impgroup.org .

Dyer, W. Gibb Jr., and Alan L. Wilkins. 1991. Better stories, not better constructs, to generate better theory: A rejoinder to Eisenhardt. Academy of Management Review 16 (3): 613–619.

Easton, Geoff. 2002. Marketing: A critical realist approach. Journal of Business Research 55 (2): 103–109.

Elliott, R., and N. Jankel-Elliott. 2003. Using ethnography in strategic consumer research. Qualitative Market Research: An International Journal 6 (4): 215–223.

Evers, W., S. Marroun, and L. Young. 2014. Using industry workshops to create idea networks for business model evolution. In 25th IMP Conference, Kedge Business School, Bordeaux, France , September.

Evers, W., S. Marroun, and L. Young. 2017. A pluralistic, longitudinal method: Using participatory workshops, interviews and lexicographic analysis to investigate relational evolution. Industrial Marketing Management 61: 182–193

Fereday, J., and E. Muir-Cochrane. 2008. Demonstrating rigor using thematic analysis: A hybrid approach of inductive and deductive coding and theme development. International Journal of Qualitative Methods 5 (1): 80–92.

Friedrichs, J., and F. Kratochwil. 2009. On acting and knowing: How pragmatism can advance international relations research and methodology. International Organization 63 (04): 701–731.

Glesne, C., and A. Peshkin. 1992. Becoming qualitative researchers . White Plains, NY: Longman.

Green, J., K. Willis, E. Hughes, R. Small, N. Welch, L. Gibbs, and J. Daly. 2007. Generating best evidence from qualitative research: The role of data analysis. Australian and New Zealand Journal of Public Health 31 (6): 545–550.

Guba, E.G., and Y.S. Lincoln. 1994. Competing paradigms in qualitative research. Handbook of Qualitative Research 2: 163–194.

Gummesson, Evert. 2003. All research is interpretive! Journal of Business & Industrial Marketing 18 (6/7): 482–492.

Healy, M., and C. Perry. 2000. Comprehensive criteria to judge validity and reliability of qualitative research within the realism paradigm. Qualitative Market Research: An International Journal 3 (3): 118–126.

Järvensivu, T., and J.Å. Törnroos. 2010. Case study research with moderate constructionism: Conceptualization and practical illustration. Industrial Marketing Management 39 (1): 100–108.

Kass, R.E., and A.E. Raftery. 1995. Bayes factors. Journal of the American Statistical Association 90 (430): 773–795.

Leximancer Manual Version 4. 2011. www.leximancer.com .

Madill, A., A. Jordan, and C. Shirley. 2000. Objectivity and reliability in qualitative analysis: Realist, contextualist and radical constructionist epistemologies. British Journal of Psychology 91 (1): 1–20.

Marroun, S., and L. Young. 2015. The motivations and behaviours of professional networking. In 26th IMP Conference, University of Southern Denmark, Kolding, Denmark , August.

Mauthner, N.S., and A. Doucet. 2003. Reflexive accounts and accounts of reflexivity in qualitative data analysis. Sociology 37 (3): 413–431.

Mays, N., and C. Pope. 1995. Qualitative research: Rigour and qualitative research. BMJ 311 (6997): 109–112.

McGhee, G., G.R. Marland, and J. Atkinson. 2007. Grounded theory research: Literature reviewing and reflexivity. Journal of Advanced Nursing 60 (3): 334–342.

Miles, Matthew B., and A. Michael Huberman. 1994. Qualitative data analysis: An expanded sourcebook . USA: Sage.

Munksgaard, K.B. 2015. Is this network for you or for me? The pursuit of self and collective interests in a strategic network. Journal of Business and Industrial Marketing 30 (3/4): 279–289.

Munksgaard, K.B., and C.J. Medlin. 2014. Self- and collective-interests: Using formal network activities for developing firms’ business. Industrial Marketing Management 43 (4): 613–621.

Patton, Michael Quinn. 1987. How to use qualitative methods in evaluation . USA: Sage.

Peter, J. Paul, and Jerry C. Olson. 1983. Is science marketing? Journal of Marketing 47 (4): 111–125.

Pratt, M.G. 2008. Fitting oval pegs into round holes tensions in evaluating and publishing qualitative research in top-tier North American journals. Organizational Research Methods 11 (3): 481–509.

Ritchie, J., and L. Spencer. 2002. Qualitative data analysis for applied policy research. The Qualitative Researcher’s Companion 573: 305–329.

Riley, R., and L. Love. 1999. The state of qualitative tourism research. Annals of Tourism Research 27 (3): 164–187.

Rong, Baiding, and Ian F. Wilkinson. 2011. What do managers’ survey responses mean and what affects them? The case of market orientation and firm performance. Australasian Marketing Journal (AMJ) 19 (3): 137–147.

Silverman, D. 2007. A very short, fairly interesting and reasonably cheap book about qualitative research . London: Sage.

Smith, Andrew, and Michael Humphreys. 2006. Evaluation of unsupervised semantic mapping of natural language with Leximancer concept mapping. Behavior Research Methods 38 (2): 262–279.

Strauss, Anselm, and Juliet M. Corbin. 1997. Grounded theory in practice . USA: Sage.

Strahle, W., and R.L. Spiro. 1986. Linking market share strategies to salesforce objectives, activities, and compensation Policies. Journal of Personal Selling and Sales Management , 6, 11–18.

Thompson, Michael, and Louise Young. 2014. Measuring complex patterns in space-time. Australasian Marketing Journal (special issue on Complexity and Agent-based Modeling), 22 (1): 28–35. ISSN 1441–3582.

Timmermans, S., and I. Tavory. 2012. Theory construction in qualitative research from grounded theory to abductive analysis. Sociological Theory 30 (3): 167–186.

Uslay, C., N.K. Malhotra, and F.C. Allvine. 2006. Predatory pricing and marketing theory: Applications in business-to-business context and beyond. Journal of Business-to-Business Marketing 13, 65–116.

Wilkinson, I., and L. Young. 2004. Improvisation and adaptation in international business research interviews chapter 4. In Handbook of Qualitative Research Methods for International Business , ed. R. Marschan-Piekkari and C. Welch. Cheltenham, UK and Northhampton, MA: Edward Elgar. ISBN 1 84376 083 5.

Wilkinson, Ian, and Louise Young. 2013. The past and the future of business marketing theory. Industrial Marketing Management 42 (3): 394–404.

Wolcott, Harry F. 1994 Transforming qualitative data: Description, analysis, and interpretation . USA: Sage.

Young, Louise. 2006. Trust: Looking forward and back. Journal of Business and Industrial Marketing 21 (7): 439–445.

Young, Louise, and Nigel Bairstow. 2012. Narrative event methods: Understanding how business market processes equilibrate and change over time. In Proceedings, Industrial Marketing and Purchasing Conference . www.impgroup.net .

Young, L., and K. Daniel (2003) Affectual trust in the workplace. International Journal of Human Resource Management 14 (1): 139–155.

Young, L., and S. Denize. 2008. Competing interests: The challenge to collaboration in the public sector. International Journal of Sociology and Social policy 28 (1/2): 46–58.

Young, Louise, Sara Denize, Ekta Nankani, Simeon Simoff, and Ian Wilkinson. 2010. Researching the structures and processes of collaborative academic networks (competitive paper) proceedings, academy of marketing science annual conference May 26–May 29 Portland . USA: Oregon.

Young, L., and L. Freeman. 2008. A case for contrast as a catalyst for change. International Journal of Learning 15 (3): 295–304.

Young, Louise, Ian Wilkinson, and Andrew Smith. 2015. A scientometric analysis of publications in the journal of business-to-business marketing 1993–2014. Journal of Business-to-Business Marketing 22 (1–2): 111–123.

Zhang, Y., and Wildemuth, B.M. 2009. Qualitative analysis of content. In Applications of social research methods to questions in information and library science , 308–319.

Download references

Acknowledgements

We would like to thank our co-researchers, Tamsin Angus-Leppan, Sue Benn, Christine Burton, Kerry Daniels, Sara Denize, Winie Evers, Sana Marron, Chris Medlin, Andrew Smith and Ian Wilkinson for the work done with us on the projects presented in our examples. A special acknowledgement goes to the developer of Leximancer, Andrew Smith, for his amazing vision in developing Leximancer and tremendous help over an extended period of research.

Author information

Authors and affiliations.

Western Sydney University, Sydney, Australia

Louise Young

University of Southern Denmark, Kolding, Denmark

Louise Young & Kristin B. Munksgaard

You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Kristin B. Munksgaard .

Editor information

Editors and affiliations.

Per Vagn Freytag

Ranked concept lists for five research phases of Advertising Project (presented in Evers et al. 2017 ).

Phase, interview 1			Phase 2, workshop 1			Phase 3, interview 2			Phase 4, workshop 2		Phase 5, workshop 3

	141	74		66	90		130	100		27		102	100
	141	74		52	71		110	85		26		85	83
							28	22				23	23

kunder (customers)	191	100	kunde (customer)	73	100	kunder (customers)	117	90	opgave (task)	70	måde (the way)	94	92
opgave (task)	130	68	vores (our)	48	66	se (see)	106	82	kunde (customer)	63	opgave (task)	90	88
forhold (relation)	100	52	opgave (task)	47	64	forhold (relation)	103	79	forhold (relation)	61	kunde (customer)	84	82
tid (time)	89	47	sammen (together)	46	63	sammen (together)	84	65	tillid (trust)	43	forhold (relation)	81	79
virksomhed (company)	85	45	virksomhed (company)	42	58	store (large)	74	57	sammen (together)	42	sammen (togeteher)	54	53
vores (our)	75	39	tid (time)	38	52	arbejde (work)	61	47	netvaerk (network)	40	huske (remember)	52	51
se(see)	75	39	forhold (relation)	37	51	virksomhed (company)	53	41	hjertet (the heart)	39	giver (gives)	39	38
sammen (together)	68	36	finde (find)	31	42	vigtigt (important)	47	36	store (large)	38	se (see)	36	35
store (large)	58	30	se(see)	31	42	finde (find)	43	33	vej (pathway)	37	finde (find)	35	34
arbejde (work)	49	26	arbejde (work)	29	40	vores (our)	43	33	partnere (partners)	35	tid (time)	33	32
del (share)	44	23	relation (relation)	27	37	opgaver (task)	42	32	arm (arm)	32	rigtige (right)	29	28
indtryk (impression)	43	23	handler (act, trades)	27	37	bruge (use)	41	32	spaendende (exciting)	26	vigtige (important)	27	26
fald (decrease)	42	22	udfordringer (challenges)	24	33	tid (time)	41	32	se (see)	24	meaning (opinion, meaning)	25	25
virksomheden (the firm)	41	21	store (large)	23	32	giver (gives)	37	28	giver (gives)	22	tiltælde (coincidence)	25	25
forskellige (different)	41	21	for (before)	21	29	prøve (try)	34	26	arbejde (work)	22	samarbejdspartnere (parteners)	24	24
prove (try)	41	21	lose (solve)	20	27	maerke (sense)	34	26	rigtige (proper)	21	del (share)	24	24
finde (find)	37	19	giver (gives)	20	27	del (share)	33	25	bedre (better)	19	sker (happen)	24	24
vaerdi (value)	35	18	hjaelpe (support)	19	26	mulighed (opportunity)	32	25	finde (find)	17	bedre (better)	24	24
nye (new)	33	17	behov (needs)	19	26	konkret (concrete)	32	25	tid (time)	17	hjertet (the heat)	23	23
pr (pr)	32	17	medier (media)	19	26	netvaerk (network)	32	25	penge (money)	17	vej (pathway)	22	22
tilbage (back)	30	16	penge (money)	18	25	samarbejdspartnere (partners)	27	21	prøve (try)	15	før (before)	22	22
kommunikation (communication)	29	15	muligheder (opportunities)	15	21	medier (media)	25	19			samarbejde (collabration)	22	22
lille (small)	26	14									tilbage (back)	22	22
											større (larger)	21	21
											prøve (try)	19	19
											bruge (use)	19	19
											forskellige (different)	19	19

Rights and permissions

Reprints and permissions

Copyright information

About this chapter

Young, L., Munksgaard, K.B. (2018). Analysis of Qualitative Data: Using Automated Semantic Analysis to Understand Networks of Concepts. In: Freytag, P., Young, L. (eds) Collaborative Research Design. Springer, Singapore. https://doi.org/10.1007/978-981-10-5008-4_11

Download citation

DOI : https://doi.org/10.1007/978-981-10-5008-4_11

Published : 28 September 2017

Publisher Name : Springer, Singapore

Print ISBN : 978-981-10-5006-0

Online ISBN : 978-981-10-5008-4

eBook Packages : Business and Management Business and Management (R0)

Share this chapter

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Publish with us

Policies and ethics

Find a journal
Track your research

arXiv's Accessibility Forum starts next month!

Help | Advanced Search

Computer Science > Machine Learning

Title: enhancing neural network interpretability through conductance-based information plane analysis.

Abstract: The Information Plane is a conceptual framework used to analyze the flow of information in neural networks, but traditional methods based on activations may not fully capture the dynamics of information processing. This paper introduces a new approach that uses layer conductance, a measure of sensitivity to input features, to enhance the Information Plane analysis. By incorporating gradient-based contributions, we provide a more precise characterization of information dynamics within the network. The proposed conductance-based Information Plane and a new Information Transformation Efficiency (ITE) metric are evaluated on pretrained ResNet50 and VGG16 models using the ImageNet dataset. Our results demonstrate the ability to identify critical hidden layers that contribute significantly to model performance and interpretability, giving insights into information compression, preservation, and utilization across layers. The conductance-based approach offers a granular perspective on feature attribution, enhancing our understanding of the decision-making processes within neural networks. Furthermore, our empirical findings challenge certain theoretical predictions of the Information Bottleneck theory, highlighting the complexities of information dynamics in real-world data scenarios. The proposed method not only advances our understanding of information dynamics in neural networks but also has the potential to significantly impact the broader field of Artificial Intelligence by enabling the development of more interpretable, efficient, and robust models.

Comments:	16 pages, 10 figures
Subjects:	Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
Cite as:	[cs.LG]
	(or [cs.LG] for this version)
	Focus to learn more arXiv-issued DOI via DataCite

Submission history

Access paper:.

Other Formats

References & Citations

Google Scholar
Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

Semantic Segmentation Method Based on Transformer for Autonomous Driving

14 Pages Posted: 27 Aug 2024

Zheng Liang

Inner Mongolia University

Semantic segmentation is a crucial research topic in intelligent automotive environment perception. Traditional Convolutional Neural Network (CNN) based semantic segmentation, despite being effective, often suffers from computational inefficiencies due to dense pixel-level predictions, which may hinder real-time applications. This study proposes an innovative, real-time approach to semantic segmentation in autonomous driving scenarios, leveraging Visual Transformers. These transformers utilize a self-attention mechanism, allowing for a global understanding that enhances pixel representation within the overall scene. Addressing the need for rapid and precise interpretation of diverse elements in driving environments, our method balances operational speed and accuracy, essential for autonomous vehicles. Evaluations conducted on the DriveSeg dataset demonstrate that our approach outperforms conventional semantic segmentation techniques in real-time performance, while maintaining comparable accuracy levels. This work not only underscores the efficiency of semantic segmentation but also suggests a promising direction for real-time processing in autonomous driving systems. Future research will focus on integrating this method with other perception tasks to enhance the robustness of autonomous vehicle perception systems.

Keywords: real-time, visual transformer, semantic segmentation, autonomous driving

Suggested Citation: Suggested Citation

Inner Mongolia University ( email )

Huhhot, Inner Mongolia China

Do you have a job opening that you would like to promote on SSRN?

Paper statistics, related ejournals, artificial intelligence ejournal.

Subscribe to this fee journal for more curated articles on this topic

IMAGES

4. Methodical development of semantic network analysis
Semantic network analysis procedure.
Figure A1. The result of semantic network analysis for FDI.
Semantic network analysis procedure.
Example of semantic network analysis.
Semantic network analysis diagram of key factors affecting unsafe acts

VIDEO

Network Analysis Made Simple- Eric Ma, Mridul Seth
Semantic Coding [1/10]
Semantics -Componential Analysis and Semantic Roles
Coding Task
Content Matters: Uncovering Semantic Networks and Hashtag Networks in NodeXL
Instance-Aware Semantic Segmentation via Multi-Task Network Cascades

COMMENTS

What Constitutes Semantic Network Analysis? A Comparison of Research
Semantic network analysis in communication research (Doerfel, 1998; Segev, 2021) can be traced to an analysis of word cooccurrences across posts on the first social media, Computer Bulletin Boards ...
Semantic Network Analysis as a Method for Visual Text Analytics
This paper proposes an approach on a method for visual text analytics to support knowledge building, analytical reasoning and explorative analysis. For this purpose we use semantic network models that are automatically retrieved from unstructured text data using a parametric k -next-neighborhood model. Semantic networks are analyzed with ...
Semantic network analysis (SemNA): A tutorial on preprocessing ...
Psychol Methods. 2023 Aug;28(4):860-879. doi: 10.1037/met0000463. Epub 2021 Dec 23. Authors Alexander P ... Using real-world data, we present a start-to-finish pipeline from raw data to semantic network analysis results. This article aims to provide resources for researchers, both the unfamiliar and knowledgeable, that reduce some of the ...
Semantic network analysis (SemNA): A tutorial on preprocessing
This article offers a comprehensive semantic network analysis pipeline (preprocessing, estimating, and analyzing networks), and an associated R tutorial that uses a suite of R packages to accommodate the pipeline. To date, the application of semantic network methodologies to study cognitive processes in psychological phenomena has been limited in scope. One barrier to broader application is ...
OSF
The third package, SemNeT, provides methods and measures for analyzing and statistically comparing semantic networks. Using real-world data, we present a start-to-finish pipeline from raw data to semantic network analysis results. This article aims to provide resources for researchers, both the unfamiliar and knowledgeable, that minimize some ...
Semantic Network Analysis (SemNA): A Tutorial on Preprocessing
As Christensen and Kenett (2020) stated, the application of semantic network analysis in academic research is currently limited. One impediment to wider adoption is a lack of resources for ...
Topological properties and organizing principles of semantic networks
Later, Steyvers and Tenenbaum 38 performed statistical analysis of 3 kinds of semantic networks: word associations 30, WordNet and Roget's Thesaurus 39. Apart from the above-mentioned network ...
Semantic Network Analysis in Social Sciences
The first chapters in this book offer step-by-step guidelines for conducting semantic network analysis, including choosing and preparing the text, selecting desired words, constructing the networks, and interpreting their meanings. Free software tools and code are also presented. The rest of the book displays state-of-the-art studies from ...
[PDF] Semantic Network Analysis: Techniques for Extracting
Semantic Network Analysis is a technique in which the content of a message is extracted from text and represented as a network of semantic relations between actors and issues, which can be queried to look for specific patterns and answer various research questions. ... The different perspectives of a network that are captured by each method are ...
PDF Doerfel What Constitutes Semantic Network Analysis
Semantic networks have been constructed in the following ways : (1) based on the relationship among words in a text; (2) based on traditional content analyses of text ; and (3) based on overlapping perceptions measured with scales . These various methodologies may be a function of two distinct definitions of semantic networks .
What Constitutes Semantic Network Analysis? A Comparison of Research
The authors focus on exploring the mental models of travel information providers though semantic network analysis when they market their destinations on the Internet, and provide a preliminary result for the semantic network. Expand
Semantic network analysis in consumer and marketing research
This research contributes to recent fundamental work on computational text analysis methods in consumer and marketing research (Berger et al., 2020, 2022; Humphreys and Wang, 2018) by focusing on SemNA as a computational method that centers on the visual exploration and interpretation of relationships and representations of knowledge ...
What constitutes semantic network analysis? A comparison of research
What constitutes semantic network analysis? A comparison of research and methodologies Sidebar menus. Research; Research Groups. Centers ... Doerfel, M. L. (1998). What constitutes semantic network analysis? A comparison of research and methodologies. Connections, 21(2), 16-26. Connections. 4 Huntington Street New Brunswick, NJ 08901 848-932-7500
Research trends in text mining: Semantic network and main path analysis
Semantic network and main path analysis were conducted on 1856 studies on text mining. • Using text mining as research topic or method has increased fast and widely applied. • Revealed keywords of text mining study in the 1980s and 1990s, the 2000s, the 2010s. • Identified which papers make a significant academic contribution on text mining.
Semantic Social Networks Analysis
Hybrid approaches coupling human participation with automatic means for semantic network analysis are likely to generate effective results for medium scale applications. ... (1989) Research methods in social network analysis. George Mason University Press, Fairfax. Google Scholar Galam S (2008) Sociophysics: a review of galam models. Int J Mod ...
Network text analysis: A two-way classification approach
In semantic network analysis human coding could be used to identify relations or not; in the last case the method is limited to the extrapolation and the analysis of the co-occurrence networks of words (Bullinaria & Levy, 2007) and it is also called network text analysis, that is a method where links between words in a text are encoded and ...
PDF Semantic Network Analysis as a Method for Visual Text Analytics
Abstract. This paper proposes an approach on a method for visual text analytics to support knowledge building, analytical reasoning and explorative analysis. For this purpose we use semantic ...
Textual network analysis: Detecting prevailing themes and biases in
Researchers can use this procedure as a grounded content analysis to formulate theories or as a basis to test existing hypotheses. The second part of the paper presents two studies that applied textual network analysis: (a) to identify the main themes raised by elite newspapers on the "fake news" discourse and (b) to map the topics related ...
Semantic Network Analysis as a Method for Visual Text Analytics
Abstract and Figures. This paper proposes an approach on a method for visual text analytics to support knowledge building, analytical reasoning and explorative analysis. For this purpose we use ...
Social Network Analysis
Pajek: Pajek is a program for the analysis and visualization of large networks. It's an extensive tool, offering a range of complex network metrics, and is free for non-commercial use. NodeXL: NodeXL is a free, open-source template for Microsoft Excel that allows users to display and analyze network graphs.
Analysis of Qualitative Data: Using Automated Semantic Analysis to
Qualitative research is something of the "poor relation" in mainstream academic research of markets and marketing. It has long been argued in mainstream forums that the dominant quantitative tradition enhances rigour and increases generalizability thus improving the credibility of marketing science (Peter and Olsen 1983).Most quantitative analysis is underpinned by this positivist ...
Semantic Social Networks: A Mixed Methods Approach to Digital
Responding to calls for multimodal and network ethnography integrating anthropological participant observation with digital research methods, our ethnographic method maps online conversations to analyze community meaning-making practices (Dicks et al. 2006; Howard 2002; Murthy 2008).Following Rheingold (2000, pp. xx), we define online communities as "social aggregations that emerge from the ...
[2408.14681] Enhancing Neural Network Interpretability Through
The Information Plane is a conceptual framework used to analyze the flow of information in neural networks, but traditional methods based on activations may not fully capture the dynamics of information processing. This paper introduces a new approach that uses layer conductance, a measure of sensitivity to input features, to enhance the Information Plane analysis. By incorporating gradient ...
Semantic Segmentation Method Based on Transformer for Autonomous ...
Abstract. Semantic segmentation is a crucial research topic in intelligent automotive environment perception. Traditional Convolutional Neural Network (CNN) based semantic segmentation, despite being effective, often suffers from computational inefficiencies due to dense pixel-level predictions, which may hinder real-time applications.
Research method; 'SNA' refers to semantic network analysis, the
Together, the research under these topics and their keywords forms a massive part of CER, and is well aligned with the keyword analysis of top research trends in Section 2.2, as well as findings ...
Full article: Integrating trajectory data and demographic
Recently, a social-sensing method that uses social media data attracted much attention, as it enables cost-effective and streamlined data collection, distinguishing itself from conventional methodologies. At present, social sensing is becoming a research method of spatial interaction analysis (Y. Liu et al. Citation 2015).

Save citation to file

Add to My Bibliography

Semantic network analysis (SemNA): A tutorial on preprocessing, estimating, and analyzing semantic networks

Similar articles

LinkOut - more resources

Main content

Add new wiki page

Page permissions have changed

Wiki page deleted

Connected to the collaborative wiki

Connecting to the collaborative wiki

Collaborative wiki is unavailable

Browser unsupported

Start managing your projects on the OSF today.

Topological properties and organizing principles of semantic networks

Similar content being viewed by others

Topological methods for data modelling

Structural differences in the semantic networks of younger and older adults

Embedding the intrinsic relevance of vertices in network analysis: the case of centrality metrics

Related work

General properties of semantic networks

Connectedness

Degree distribution

Degree assortativity

Clustering coefficient

Language-specific properties

Language classification

Overview of semantic networks from eleven languages

Language inflection

Part-of-speech tags

Merging of word inflections

The number of inflections

Similarity and complementarity in semantic networks

Data Availibility

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding author

Ethics declarations

Additional information

Supplementary Information

About this article

Share this article

Quick links

Get Citation

TABLE OF CONTENTS

What Constitutes Semantic Network Analysis? A Comparison of Research and Methodologies'

Tables from this paper

158 Citations

The Pivotal Role of AJC in the Growth of Communication Research in Asia: A semantic network analysis

A Semantic Network Analysis of the International Communication Association

Semantic network analysis in consumer and marketing research: application areas in phygital contexts

Design/methodology/approach

Research limitations/implications

Practical implications

Originality/value

Introduction

Theoretical background

Approaches to semantic network analysis in consumer and marketing research

Application areas for semantic network analysis in phygital contexts

Application area 1: investigating phygital experiences and consumption phenomena

Sample and procedure.

Discussion application area 1.

Application area 2: exploring phygital consumer and market discourse, trend and practices

Discussion application area 2.

Application area 3: capturing phygital social constructs

Discussion application area 3.

Limitations and future research

Acknowledgements

Corresponding author

Related articles

Report an issue or find answers to frequently asked questions

Social Network Analysis – Types, Tools and Examples

Social Network Analysis

Types of Social Network Analysis

Whole Network Analysis

Ego Network Analysis

Two-mode (or Bipartite) Network Analysis

Dynamic Network Analysis (DNA)