An overlapping semantic community detection algorithm base on the ARTs multiple sampling models

Abstract

Since the Semantic Social Network (SSN) is a new kind of complex networks, the traditional community detection algorithms require giving the number of the communities and could not detect the overlapping communities. To solve this problem, we propose improving multiple sampling models ARTs, consisting of ART, LART, ARTF and LARTF, sampling the textual information specific to node, link, node field, and link field correspondingly. The proposed ARTs models separate the semantic community detection into context sampling and communities detecting stage. After the context sampling, the quantized semantic coordinate is allocated to each sampling element, by which the cohesion for each sampling field can be established, avoiding the presetting of the number of communities. As the ARTs models are not easy to convergence, we explore the multiple sampling to accelerate the convergence, and the parameters of ARTs are analyzed by experimental analysis. In evaluation aspect, some traditional evaluation models are extended for semantic community measurement. Finally, efficiency of ARTs is verified by experiment.