Analysing Topic Networks

The Power of Bigrams

A key technique in topic network analysis is the use of bigrams — pairs of words that frequently appear together. Traditional topic models often rely on unigrams (single words), but these can fail to capture the nuances of language, especially when two words are commonly paired (e.g., “marketing analytics”). Bigrams can significantly enhance the clarity and specificity of topics.

For example, if two terms co-occur frequently in posts (e.g., “marketing analytics” and “quantitative research”), they form a stronger connection in the network. Bigrams are discussed in greater detail in Section N-grams, Bigrams and Trigrams.

Centrality Measures

Centrality Measures are used to identify key topics in a network:

Centrality-Measures
• Degree Centrality: Represents topics frequently mentioned or linked to others.
• Closeness Centrality: Highlights topics that act as important hubs, spreading information quickly.
• Betweenness Centrality: Identifies topics that connect different sections of the network.
• Eigenvector Centrality: A measure of a topic’s importance, based on how connected it is to other influential topics.

Using Centrality to Understand Influence in Social Networks

A notable application of centrality measures is PageRank, a concept popularised by Google to rank web pages based on their importance. In the context of social network analysis, PageRank can be applied to topics, assigning higher scores to those frequently referenced by other influential topics.

By analysing these centrality measures, companies and researchers can identify which topics dominate a conversation, serve as key influencers, or are highly interconnected with other important themes.