Community Detection Based on Social Network Analysis in Question and Answer Systems

A community-based question and answer (CQA) system is an integrated Internet platform for users to share knowledge, and it becomes very popular in recent years. In this paper, we study social network structures of CQA systems based on the dataset collected from “Baidu Knows”, which is the largest CQA system in China. By exploiting the question-answer interactions among the users, we construct two networks, and find that both exhibit strong social network characteristics, including scale-free, smallworld, and rich-club properties; in addition, interest-oriented user communities can be detected from the networks. Motivated by the observation, we propose a novel algorithm, namely multilayer speakerlistener label propagation algorithm (MSLPA), to detect user communities. Evaluation using real-world data from “Baidu Knows” suggests that the algorithm can effectively detect the genuine communities in which users share common interests. Comparing with existing approaches, our algorithm is able to avoid forming large number of tiny communities and integrates similar communities. We believe that with these nice properties, the proposed algorithm can be applied in many applications, such as question/answer recommendation, for improving the knowledge qualities in the CQA systems.