Semantic Network Analysis in Social Sciences

Elad Segev

• Publisher: Routledge
• Year: 2021
• Print ISBN: 9780367636500
• eBook ISBN: 9781000471915
• Edition: 1
• Topic: Psychology
• Subtopic: Research & Methodology in Psychology
• Index: Psychology

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.

1

How to conduct semantic network analysis

Elad Segev

DOI: 10.4324/9781003120100-2

The process of semantic network analysis generally consists of two main stages: text preparation and analysis. First, researchers are required to choose the right text and prepare it for analysis. Some texts, such as instant messages in a chat group, may be less informative, and the content of this text would therefore be of no interest. Other texts may not fit the research questions. The corpus should be well defined and properly prepared for analysis, including the selection of desired words to be part of the network analysis. In the second stage, the network is constructed, analyzed, and interpreted. This stage includes simplifying the network structure and identifying the clusters and the most central words in the network, which represent the main discourse and narratives appearing in the text. The following steps are recommended to be taken as part of the semantic analysis:

Stage 1 Preparing your corpus for analysis

1.1 Defining and refining the sample

The first stage requires a clear definition of the corpus to be studied and its relevance to the research question. When researchers have a clear question in mind, it is sometimes sufficient to skim through the text to see if the answer might be hiding in it. When research questions are not yet formulated, skimming through the text enables researchers to assess whether there are interesting issues that could be raised and highlighted. As a rule of thumb, a good text is a provocative text. It is a text that contains debates, tensions, contradictions, biases, explicit, or implicit agendas. It should be rich and diverse enough in terms of vocabulary and the span of topics to allow a meaningful analysis. Thus, for example, literature and art, news articles, political speeches, social media debates on Twitter or Facebook, Wikipedia articles, transcripts of interviews or films, religious scripts, and even academic papers may be appropriate for semantic network analysis. The next chapters in the book will demonstrate how to analyze different types of texts and the strengths and limitations of such an analysis.

In the last two decades various new sources for text analysis have become available to researchers, such as Factiva and Nexis Uni databases for news, legal, government, and business information, social media content on Twitter and Facebook API, blogs and Wikis, search engine results, and so forth. Together with new sources, social scientists rely also on traditional sources such as transcripts of in-depth interviews and collections of research articles. As the amount of data increases, semantic network analysis offers an innovative and fresh approach for both new and more traditional types of data, as it allows researchers to organize, visualize, and obtain an initial map of the main topics and themes appearing in those corpora.

1.2 Selecting the words to be analyzed

The second stage is crucial as it dramatically affects the rest of the analysis. In this stage researchers need to choose the keywords, expressions, or phrases to be studied in light of the theory and research questions. In traditional discourse analysis researchers closely read the text to identify those key terms. Reading through the text is still a very important stage in semantic network analysis, as it helps knowing better the topics, use of words, and the communication dynamic. It also helps in revealing central words that appear in specific contexts and may have essential cultural and social meanings. Thus, for example, when reading through political speeches, one may find the word “I” to be very important, as it reflects the rhetoric and position of the speaker (see Chapter 5). In many other types of text the word “I” is merely an unnecessary stop-word that could be ignored.

Together with these qualitative methods, a common computer-based approach would be to construct a list of frequent words in the text and choose the top 100–200 words. This can be done with the help of freely available online tools (see Table 1.1). Another option would be to focus on specific words such as names of people, companies, organizations, or countries mentioned in the text. This could help to reveal the dynamic between the actors involved in the discourse. In this stage researchers often combine top-down and bottom-up approaches. From the top-down perspective the theory, research questions, and hypotheses should always be in mind when choosing the relevant keywords. From the bottom-up perspective, researchers should be open to including keywords that add new categories and fresh dimensions to the analysis.

Figure 1.1 demonstrates how to extract the most frequent words in the first chapter of the Bible, using the word frequency calculator available online from Segev’s (2020) set of text analysis tools.

Figure 1.1 Frequent words from the first chapter of the Bible.

It is important to note, however, that before constructing a list of the most frequent words the text might need to undergo some initial processing and preparation. In most languages, words are separated by spaces, which enables algorithms to automatically detect and count them as separate units. In some languages, such as Chinese or Japanese, researchers should first use software to separate words from each other (a process known as “tokenization”, see also the examples in Chapters 3 and 7).

1.3 Cleaning and refining the list of words

Apart from commonly used words (pronouns, prepositions, and conjunctions such as “the”, “that”, or “of”, which are known as “stop-words”), there are often frequent words in the text (such as “one” or “able”) that may appear in different contexts and have little or no relevance to the theory and research question. When constructing networks of words co-occurring in the text, these general words often appear at the center, and thus prevent from identifying the overall structure of the network and the existence of clusters of words. They should be therefore removed from the final list before proceeding to the next stage.

Another important action to be taken at this stage would be merging families of derivationally related words with similar meanings (also known as “stemming” and “lemmatization”). For example, conjugations (e.g., “believe” and “believing”), single and plural forms (e.g., “president” and “presidents”), or prepositions that, in some languages, can be part of the same word, should be merged. Similarly, common expressions such as “social media” or “united states” should be combined.

One can decide to create several lists, each with a focus on different aspects of the text. Thus, for example, selecting nouns and expressions could be useful when attempting to identify the most prevailing topics (Rule et al., 2015). The focus on adjectives could be particularly useful when the research aim is to study sentiments, opinions, and views, while verbs are useful to highlight actions.

Stage 2 Constructing and analyzing the network

Converting the text and list of words to a network. The unique power of semantic network analysis is its focus on the interactions between words and their co-occurrences to produce meaning. Once the clean list of about 100–200 words has been finalized, it is possible to generate a link list or a matrix of word pairs that co-occur together. For this purpose, one should first define the “window” (such as a sentence, a post, a tweet, a paragraph, or even a news item), in which each pair of words should be counted. Some free online tools (see Table 1.1) enable converting any text to a link list format given the desired list of predefined words. While words act as the nodes in the network, the ties between them are the number of sentences, posts, paragraphs, or news items in which they appear together.

Figure 1.2 demonstrates how to use Segev’s (2020) set of text analysis tools to convert a list of desired frequent words, appearing at least twice in the first chapter of the Bible, into a link list. This link list contains word pairs ...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Dedication
• Table of Contents
• List of figures
• List of tables
• Acknowledgements
• List of contributors
• Introduction
• 1. How to conduct semantic network analysis
• 2. The news coverage of threats: Iranian nuclear programs in Israeli press
• 3. Provocation narratives in Chinese and US newspapers
• 4. Cable news channels’ partisan ideology and market share growth as predictors of social distancing sentiment during the COVID-19 pandemic
• 5. Politicizing the Holocaust: A comparative analysis of Israeli and German speeches
• 6. Network of cleavages?: British paradiplomacy in the (digital) international discourse around Brexit
• 7. Sexual assaults blindsided by politics on Twitter: Semantic Network analysis of symbolic #MeToo cases in Japan and South Korea
• 8. Time to be happy: WhatsApp and phatic communication within the extended family
• 9. School improvement: Semantic network analysis of the literature
• 10. Identifying patterns in communication science: Mapping knowledge structures using semantic network analysis of keywords
• 11. Summary and conclusion
• Glossary
• Index