Two-steps graph-based collaborative filtering using user and item similarities: Case study of E-commerce recommender systems

Abstract

Collaborative filtering has been used extensively in the commercial recommender system because of its effectiveness and ease of implementation. Collaborative filtering predicts a user's preference based on preferences of similar users or from similar items to items that are purchased by this user. The use of either user-based or item-based similarity is not sufficient. For that particular issues, hybridization of user-based and item-based in one collaborative filtering recommender system can be used to sort relevant item out of a set of candidates. This method applies similarity measures using link prediction to predict target item by combining user similarity with item similarity. The experiment results show that the combination of user and item similarities in two-steps collaborative filtering setting improves accuracy compared to the algorithm applying only user or item similarity.