Nonparametric Models for Characterizing the Topical Communities in Social Network

Ziqi Liu,Qinghua Zheng,Fei Wang,Buyue Qian

DOI: https://doi.org/10.1016/j.neucom.2016.07.053

IF: 6

2016-01-01

Neurocomputing

Abstract:Topical communities have shown useful tools for characterizing social networks. However data in social networks often come as streams, i.e., both text content (e.g., emails, user postings) and network structure (e.g., user friendship) evolve over time. We propose two nonparametric statistic models where infinite latent community variables coupled with infinite latent topic variables. The temporal dependencies between variables across epochs are modeled via a rich-gets-richer scheme. We focus on characterizing three dynamic aspects in social streams: the number of communities or topics changes (e.g., new communities or topics are born and old ones die out); the popularity of communities or topics evolves; the semantics such as community topic distribution, community participant distribution and topic word distribution drift. Furthermore, we develop an effective online posterior inference algorithm for the models, which is concordant with the online nature of social streams. Experiments using real-world data show the effectiveness of our model at discovering the dynamic topical communities in social streams.