Social Recommendation Based On Implicit Friends

In the era of big data,conventional recommender systems suffer from two serious problems: data spasity and cold-start.With the popularity of social platforms,an increasing amount of social relations are being generated.Intuitively,online social relations can be incorporated into recommender systems to mitigate the problems of data sparsity and cold-start.Based on this intuition,social recommender systems emerged and have drawn a lot of attention.However,social recommender systems are found not as effective as expected when being deployed.The primary negative finding is that explicit social relations are highly noisy.To this end,the discovery of reliable user relations plays a central role in advancing social recommender systems.Existing social recommender systems are mainly based on explicit social relations.A few systems which based on reliable relations only pay attention to the social network and ignore the potential value in user-item bipartite.This paper proposes a novel social recommendation framework to identify implicit friends toward the discovery of more credible user relations.In particular,implicit friends are those who share similar tastes but could be distant from each other on the network topology of social relations.Methodologically,to find the implicit friends for each user,this paper first model the whole system as a heterogeneous information network,and then leverage the meta-paths to capture the similarity of users through embedding representation learning.In this way,it enables us to alleviate adverse consequences of unreliable social relations for more effective recommendation.Experimental analysis on three realworld datasets demonstrates that the proposed method outperforms other state-of-the-art baselines.The main contributions of this paper are as follows:· this paper systematically examines the bottlenecks and limitations of existing social recommender systems.· this paper formally formulates the user-item bipartite network and user-user social network as a heterogeneous information network,which enriches the existing measure of user interactions,and shows how the implicit friends are accurately identified over the heterogeneous information networks by carefully designed meta-paths and embedding representation learning.· this paper formally introduces the concept of implicit friends for social recom-mendation and proposed a new social recommendation method which adaptively incorporates the implicit friends into social Bayesian Personalized Ranking to enhance the recommendation performance.· this paper rigorously conducts experiments to validate the effectiveness of the proposed framework in recommending items and explain why the implicit friends can improve social recommendation.