Two applications of clustering techniques to twitter: Community detection and issue extraction

Abstract

Twitter's recent growth in the number of users has redefined its status from a simple social media service to a mass media. We deal with clustering techniques applied to Twitter network and Twitter trend analysis. When we divide and cluster Twitter network, we can find a group of users with similar inclination, called a "Community." In this regard, we introduce the Louvain algorithm and advance a partitioned Louvain algorithm as its improved variant. In the result of the experiment based on actual Twitter data, the partitioned Louvain algorithm supplemented the performance decline and shortened the execution time. Also, we use clustering techniques for trend analysis. We use nonnegative matrix factorization (NMF), which is a convenient method to intuitively interpret and extract issues on various time scales. By cross-verifying the results using NFM, we found that it has clear correlation with the actual main issue. © 2013 Yong-Hyuk Kim et al.