Multi-type Relational Data Clustering for Community Detection by Exploiting Content and Structure Information in Social Networks

Abstract

Social Networks popularity has facilitated the providers with an opportunity to target specific user groups for various applications such as viral marketing and customized programs. However, the volume and variety of data present in a network challenge the identification of user communities effectively. The sparseness and heterogeneity in a network make it difficult to group the users with similar interests whereas the high dimensionality and sparseness in text pose difficulty in finding content focused groups. We present this problem of discovering user communities with common interests as the multi-type relational data (MTRD) learning with the content and structural information, and propose a novel solution based on non-negative matrix factorization with added regularization. We empirically evaluate the effectiveness of the proposed method on real-world Twitter datasets benchmarking with the state-of-the-art community discovery and clustering methods.