Topic Analysis to Identify Communities

Abstract

Abstract Being able to detect communities in social networks can be an aid in understanding trends, assist moderation efforts and build recommendation systems. In this paper we explore the use of topic models for community detection by proposing two such models, LDAC and LDACS, based off of Latent Dirichlet Allocation (LDA) [1] and the Community Topic Model [8]. These models are compared to LDA and evaluated on datasets collected from Twitter and Reddit. It is concluded that LDACS may be a reasonable and simple model for community detection, but with further study needed, and that LDAC gives some credence to utilizing both topics and communities in a model, but does itself not produce sufficient results to weigh up for its complexity, although training it on more data might remedy this.