Discovering Communities with SGNS Modelling-based Network connections and Text communications Clustering

By the community discovery, the microblogging services facilitate diverse applications such as viral marketing, disaster management, customized programs, and many more. However, the sparseness and heterogeneity of user networks and text content make it difficult to group users with a similar interest. In this paper, we present a novel method to discover user communities with common interests. The proposed method utilizes both text content and interaction network information where network information is modeled using the concept of Skip-Gram with Negative Sampling for Non-negative Matrix Factorization. Empirical analysis using several real-world Twitter datasets shows that the proposed method is able to produce accurate user communities as compared to the state-of-the-art community discovery and clustering methods.