Probabilistic Models for Recommendation in Social Networks

Recommender systems are becoming tools of choice to select the online information relevant to a given user. Collaborative filtering is the most popular approach to building recommender systems and has been successfully employed in many applications. However, collaborative filtering based approaches perform poorly for so-called cold start users. With the advent of online social networks, the social network based approach to recommendation has emerged. This approach assumes a social network among users and makes recommendations for a user based on the ratings of the users that have direct or indirect social relations with the given user. As one of their major benefits, social network based approaches have been shown to reduce the problems with cold start users. In this research we propose novel methods to address the recommendation problem in online social networks. To better understand the underlying mechanisms of user behavior in a social network, we first propose a model to capture the temporal dynamics of user behavior based on different effects influencing the behavior of users in rating items and creating social relations (e.g. social influence, social selection and transitivity of relations). Then we propose a memory based approach based on random walk models to perform recommendation in social networks. Matrix factorization is the most prominent model based approach for collaborative recommendation. We extend matrix factorization and propose a model that takes into account the social network as well as the rating matrix. Finally, we present a mixed membership community based model for recommendation in social networks based on stochastic block models. This model is capable of performing both rating and link prediction. All methods have been experimentally evaluated and compared against state-of-the-art methods on real life data sets from Epinions.com, Flixster.com and Flickr.com. The Flixster data set has been crawled and published as part of the research during this thesis. Experimental results show that our proposed models achieve substantial quality gains compared to the existing methods.