JUST-BPR: Identify Implicit Friends with Jump and Stay for Social Recommendation

Abstract

Recommender constantly suffers from the problems of data sparsity and cold-start. As suggested by social theories, people often alter their ways of behaving and thinking to cater to social environments, especially their friends. For this reason, prior studies have integrated social relations into recommender systems to help infer user preference when there are few available data, which is known as social recommendation. However, explicit social relations are also sparse and meanwhile are usually noisy. To enhance social recommendation, a few studies identify more reliable implicit relations for each user over the user-item and user-user networks. Among these research efforts, meta-paths guided search shows state-of-the-art performance. However, designing meta-paths requires prior knowledge from domain experts, which may hinder the applicability of this line of research. In this work, we propose a novel social recommendation model (JUST-BPR) with a meta-path-free strategy to search for implicit friends. Concretely, We adopt the idea of ’Jump and Stay’, which is a heterogeneous random walk technique, to social recommendation. Based on this idea, we manage to bypass the design of meta-paths and obtain high-quality implicit relations in a more efficient way. Then we integrate these implicit relations into an augmented social Bayesian Personalized Ranking model for top-N recommendation. Experiments on two real-world datasets show the superiority of the proposed method and demonstrate the differences between implicit friends discovered by meta-paths and JUST-BPR, respectively.