Social network based recommender systems

Abstract

Recommender systems have become de facto tools for suggesting items that are of potential interest to users and achieving great success in e-commerce. Many famous online vendors such as Amazon and Netix leverage recommender systems to advertise products to customers. Predicting a user's rating on an item is the fundamental recommendation task. Traditional methods that generate predictions by analyzing the user-item rating matrix perform poorly when the matrix is sparse. Recently, approaches that use data from social networks to improve the accuracy of rating prediction are emerging. However, most of the social network based recommender systems only consider direct friendships and they are less effective when the targeted user has few social connections. In this thesis, we review important rating prediction approaches in traditional and social based recommender systems. We extend SNRS, a state-of-the-art social recommender system by considering classifying the correlations between pairs of users ratings to enhance accuracy and including more users in the temporal influence links of the target user to improve the coverage. In addition, we boosted the effectiveness of social recommender systems based on matrix factorization, by proposing two models that incorporate the overlapping community regularization into the matrix factorization framework differently. Our empirical studies on real data show that our approaches outperform baselines in both traditional and social network based recommender systems.