A Hypergraph Model for Incorporating Social Interactions in Collaborative Filtering

In this paper, a new collaborative filtering method for item recommendation incorporating information from social networks is proposed. Item ratings and friendship ties are combined to define similarity among users and detect communities. Friendships and items are encoded as hyperedges of a hypergraph of users from which communities are detected. The algorithm is tested on a real-world dataset from an online recommendation platform including social network features. It is shown to outperform state-of-the-art recommendation systems both when the ratings in the training set are densely or sparsely distributed (hence dealing with the cold start problem). It also tends to recommend a greater diversity of items than traditional collaborative filtering.