Towards Recommendation Using Interest-Based Communities in Attributed Social Networks

Social networks can be modeled as attributed networks whose nodes represent users, edges represent relationships among users (e.g. friendship/follow) and attribute vectors hold properties of nodes and/or edges. In this paper, we consider friends' recommendation based on interest-based communities generated from topic based attributed social networks (TbASN). In our model, an attribute vector is not just a container for explicit users' profile data that is stored in social network's dataset, but rather holds topic vectors that are derived from analyzing the implicit interest of users' that are aggregated from his/her posts on the social network (e.g. tweets in Twitter, posts in Facebook). In our framework, topics of interest are represented as a hierarchy of topics (Topics/Subtopics) forming hierarchical interest-based communities. Users within each interest-based community are clustered according to their profile features (age, location, education etc.). Those clusters are later used in recommendations where recommendations target members of the same cluster to guarantee the quality and coherence of recommendations. In addition, we propose a recommendation selection approach to handle the large number of recommended candidates. The main advantage of the proposed approach is that it considers multiple criteria for candidate selection including the number of common communities, the resemblance in basic features, as well as network proximity. In addition to recommending friends of similar interests, frequent pattern mining is used to discover frequently occurring interests in order to be used in recommending communities for users to join. Although our approach is generic and can be applied to most of the existing social networks, we used Twitter as our target social network.