RIT: Enhancing Recommendation with Inferred Trust

Trust-based recommendation, which aims to incorporate trust relationships between users to improve recommendation performance, has drawn much attention recently. The focus of existing trust-based recommendation methods is on how to use the observed trust relationships. However, the observed trust relationships are usually very sparse in many real applications. In this paper, we propose to infer some unobserved trust relationships to tackle the sparseness problem. In particular, we first infer the unobserved trust relationships by propagating trust along the observed trust relationships; we then propose a novel trust-based recommendation model to combine observed trust and inferred trust where their relative weights are also learnt. Experimental evaluations on two real datasets show the superior of the proposed method in terms of recommendation accuracy.