Boosting Temporal Community Detection via Modeling Community Evolution Characteristics

Abstract

Community structure analysis in dynamic network is widely concerned in various fields, which mainly focus on temporal community detection and community evolution analysis. Most of the related works usually fist detect communities and then analyze evolution. This leads to a loss of evolution information on temporal community detection because block structures and evolution characteristics coexist in dynamic networks. Even thought a few modelbased approaches consider the evolution characteristics into community detection, they need to know the number of communities in advance and ignore automatic determination of the number of communities, which is a model selection problem. In the paper, we propose an model, Evolutionary Bayesian Non-negative Matrix Factorization (EvoBNMF), to model community structures with evolution characteristics for boosting the performance of temporal community detection. In detail, EvoBNMF introduces evolution behaviors, which quantify the transition relationships of communities between adjacent snapshots, to describe the evolution characteristics of community structure. Innovatively, EvoBNMF can catch the most appropriate number of communities autonomously by shrinking the corresponding evolution behaviors. Experimental results from synthetic networks and real-world networks over several state-of-the-art methods show that our approach has superior performance on temporal community detection with the virtue of autonomous determination of the number of communities.