Content Patterns in Topic-Based Overlapping Communities

Abstract EN

Understanding the underlying community structure is an important challenge in social networkanalysis.

Most state-of-the-art algorithms only consider structural properties to detect disjointsubcommunities and do not include the fact that people can belong to more than one communityand also ignore the information contained in posts that users have made.

To tackle this problem,we developed a novel methodology to detect overlapping subcommunities in online social networksand a method to analyze the content patterns for each subcommunities using topic models.

This paper presents our main contribution, a hybrid algorithm which combines two different overlappingsub-community detection approaches: the first one considers the graph structure of thenetwork (topology-based subcommunities detection approach) and the second one takes the textualinformation of the network nodes into consideration (topic-based subcommunities detectionapproach).

Additionally we provide a method to analyze and compare the content generated.

Tests on real-world virtual communities show that our algorithm outperforms other methods.