Combining topological and topical features for community detection

Abstract

Community detection is an important approach to identify community's structure in a network and can also be considered as graph clustering. This paper conducted a research about community detection using combined topological and topical features in Twitter. The combined features were compared to topological only and topical only. The topological features that were used are following-follower relationship and retweet-favorite ratio while topical features are hashtags, mentions, links and tweets. This research proposed a new node weight using retweet-favorite ratio to build topological matrix and it has been proved to have higher purity value by 30–40% and higher rand index value by 10–20%. The purity value of combining topological and topical features is also improved by 30% compared to using following-follower relationship as topological features. The highest rand index and purity values are achieved by matrix of combinied topological and topical features with multilevel community detection as clustering algorithm with 0.89 and 0.77.