Personalized Ranking in Collaborative Filtering: Exploiting l-th Order Transitive Relations of Social Ties

The use of social information in collaborative filtering is highly encouraged, as it can improve the recommendation accuracy by handling the cold start issue. The intuition of social recommendation is to reflect one's personal choice by its social neighbors. Though there exists a considerable amount of studies in this domain, no attention is paid to incorporate the transitive relationships of social ties in the ranking problem. In this paper, we exploit the lth order transitive relations of a user and extend the popular Social Bayesian Personalized Ranking (SBPR) model. The use of transitive relation creates a more granular pairwise ranking of items for a particular user and levels the user's personal choice based on the order of its social neighbors. We implement the model and conduct experiments on two real-world recommendation datasets with different values of l. We show that our model outperforms state-of-the-art pairwise ranking techniques.