Research On Social Recommender System Based On Latent Factor Model

With the rapid development of Internet technology, the cost of gathering data is becoming increasingly low. However, facing massive data, people usually end up with having numerous data but little information. It is becoming an important issue of how to provide personalized information to users based on their demands. Personalized Recommender System came into being in such a context, it focus on solving the problem of providing users with personalized information recommendation based on analyzing users’ historical behavior records, Recommender system has been widely used in many real life application after years of development.Traditional recommender system is mainly based on analyzing history record of users on items. However, with the recent rise of online social networks, more attention has been paid to the so-called Social Recommendation System which utilizing online social network to improve Recommender System. In this thesis, based on the traditional recommendation algorithms and social recommendation algorithms, we studied the trust relationship in the social connections, and proposed a trust-aware recommendation algorithm and verified the effectiveness of the algorithm through experiments. The main contribution of this thesis is as follows,(1) Proposed a trust-aware recommendation algorithm which based on Latent Factor Model. By modeling multi-faced trust, biased trust, trust transitive and mutual influence introduced by the interdependence of trust network and rating network, we gave a more accurate characterizations of the trust network and ratings network by adding both trust relationship and rating records of users’ into Latent Factor Model and combined them into a uniform optimization model.(2) The training and predicting algorithm of the proposed method are presented, including calculation of model constants, rating prediction, trust prediction, gradient calculations and model training algorithms. Besides, an adaptive learning rate updating strategy is proposed to optimize the model training process.(3) Comparative experiments are conducted on real life dataset including FilmTrust, Epinions and Ciao datasets to compare the rating prediction accuracy between the proposed method and traditional trust-aware recommendation algorithms. The experimental results show that the proposed method achieves a lower prediction error and has potential of solving the cold-start problem many recommender system failed to solve.