A MF-based Trust Prediction Model with Social and Interest Regularization in OSNs

The construction of trust plays an important role in various applications of Online Social Networks (OSNs). However, in the environment of big data, the existing trust prediction models are suffering from various problems such as the sparse social interaction and the low utilization of trust relations, which result in poor predictive performance. Meanwhile, the integration of social trust graph and recommendation system is a research hotspot which has demonstrated its effectiveness. Thus, in this paper, we propose a novel MF-based trust prediction model to complete trust relations in OSNs. Firstly, we identify authoritative trusted neighbors within the existing trust relationships and construct the regularization term of social trust effect with the limited neighbors as constraints. Then, we establish the preference similarity matrix by utilizing rating information from the recommendation systems and partition the interest community as another regularization term. Finally, a unified trust prediction model based on matrix factorization method is proposed, which integrates social trust information and interest community information. We conduct a number of experiments on two well-known datasets, Epinions and Ciao, and the results indicate that our model has better performance than the traditional SWTrust model and the MF-based Matri model. Our work is the first to introduce rating information from recommendation systems into trust prediction model and the experimental results validate its accuracy.