Friend recommendation in social networks based on multi-source information fusion

Friend recommendation (FR) in social networks has been widely studied in recent years, which mainly focuses on social relationships and user interests. Friend of Friend method is one representative. However, the disadvantage is that most of existing solutions ignored other valuable information, such as user profile, location, influence and indirect trust. In fact, being friends among users is either determined by one or two dominant factors that originate from varying information sources, or the results of multiple main factors gaming. Motivated by the observations above, we propose a scalable FR framework in social networks, where multiple sources have been integrated based on improved D-S evidence theory. More specifically, we first analyzed 7 valuable information sources and categorized them into three classes, including Personal Features, Network Structure Features and Social Features. Furthermore, we also propose a fusion recommendation framework based on D-S evidence theory which embodies the minimal conflicts among evidences. In the proposed method, we first optimize the framework by importance degree and reliability of evidence based on original D-S evidence theory. Then, we designed a novel BPA evidence function by quantifying the evidence, where each evidence measures the relevance of forming friends among users. Finally, we describe the fusion FR algorithm plugged into our recommendation framework. The experiments on real-world dataset show that our proposed approach outperforms the other state-of-the-art algorithms on five evaluation metrics. The experimental results demonstrate the effectiveness of fusing multi-source information for FR in social networks.