A trust networks recommender algorithm based on Latent Factor Model

Recommender system in social networks has been a solution to the problem of information overload, but easily suffers the challenge of data sparsity and cold start. Meanwhile, since the concept of trust has gained wide concerns, many online social networks begin to introduce trust relationship between users, which motivates us to incorporate trust information to address above problems and improve the quality of recommendation in social networks. Latent Factor Model (LFM) such as SVD++ has been proven effectiveness in social recommendation task. Therefore, this paper proposes a trust networks recommender method based on LFM for rating prediction of recommending items. Our method exploits both explicit and implicit trust information to describe users' relationship and model their social behaviors. Furthermore, we combine LFM with neighborhood model to overcome its inability of interpretation. The experimental results on Epinions can demonstrate that trust information helps alleviate the problem of data sparsity and cold start. Besides, our combination algorithm achieves higher accuracy for prediction and outperforms other trust-based algorithms, including the state-of-the-art method.