Interest before liking: Two-step recommendation approaches

Recommender systems become increasingly significant in solving the information explosion problem. Existing techniques focus on minimizing predicted rating errors and recommend items with high predicted values to people. They consider high and low rating values as liking and disliking, respectively, and tend to recommend items that users will like in the future. However, especially in the information overloaded age, we consider whether a user rates an item as a measure of interest no matter whether the value is high or low, and the rating values themselves represent the attitude to the quality of the target item. In this paper, we propose two-step recommendation approaches that recommend items matching users' interests first, and then try to find high quality items that users will like. Experiments using MovieLens dataset are carried out to evaluate the proposed methods with precision, recall, NDCG, preference-ratio and precision-like as evaluation metrics. The results show that our proposed approaches outperform the seven existing ones, i.e., UserCF, ItemCF, ALS-WR, Slope-one, SVD++, iExpand and LICF.