Reliable Potential Friends Identification Based on Trust Circuit for Social Recommendation

Abstract

Direct trust links among users may be unreliable due to noise. Simple use of these direct trust links may lead to inferior recommend effects, and most of the existed methods don’t consider the difference in trust strength. We propose a novel model called TrustE which combines the trust relationships and users similarity. Specifically, we design a new method called Trust Circuit in TrustE to model trust relationships which calculates trust values by taking into account the asymmetry, transitivity, attenuation, and multiplicity-paths of trusts. Then we calculate user similarity through meta-paths guided embedded representation learning in the heterogeneous information network. Finally, we combine trust value and users similarity to get the personalized numbers of reliable potential friends for each user and make recommendation for target user according to his friends’ preferences. The experimental results on Epinions and Douban datasets verify that TrustE is superior to other existing recommendation methods and it also has high accuracy for cold-start users’ recommendation.