A matrix factorization based trust factors model

Abstract

Considering the complexity of trust among users, we presented a new trust factor model. In this model, trust relationships among users are represented as an adjacent matrix, and trust matrix can be decomposed into two low rank dimensionality matrixes, which are trust factor matrix of users and trusted factor matrix by users, through matrix factorization in this model. User-to-user trust values are computed as inner products of corresponding vectors., Trust users can be identified according sorts of trust values. The model can give consideration to the global structure and finer-grained characteristics of trust. The experimental results show that this model has higher accuracy of trust identification and efficiency in prediction stage. The Trust factors model not only reflect users' characteristics of trust and trusted, but also can be used to explain trust propagation in trust chain.