Pairwise Learning for Coummunity Recommendation Utilizing Social Information and User-Community Interaction Degree

Nowadays, the status of social networking sites become more and more important in people’s life. Many social networking sites encourage users to create their own communities or join other’s communities to interact with other users, but there are information overload problem that users can’t easily find the communities they want to join. And this may pull users back from using the social service. In this paper, we propose a useful community recommendation approach that combine MF and LTR to model user and community’s preference, and we also incorporate both social information and user-community interactive degree in our method. The result by using a real-world dataset shows that both LTR and social information can help enhance recommendation quality evaluated by coverage and nDCG. We also show that when training pairwise learning to rank model, the recommendation quality can be further improved if one choose the trained pairs wisely. We compare some possible pair selection strategies and found that the most important thing for these pair selections is to recognize the preferable communities for a user