Hybrid Collaborative Filtering Based on Users Rating Behavior

Abstract

Several collaborative filtering (CF) approaches have been developed in order to improve the quality of the recommendations. However, this improvement has always been measured as the average quality of the performed recommendations across all the users. It has not been analyzed for each individual user. In this paper, the existence of a more precise CF approach for each user is demonstrated. So, a novel hybrid method that merges recommendations provided by different CF approaches based on a multi-class classification algorithm is proposed. This classification is performed based on the user rating behavior. Experiments have been carried out on the MovieLens and Netflix datasets. The experimental results demonstrate an improvement on quality of both predictions and recommendations using the proposed hybrid CF approach. In addition, experiments have compared state-of-the-art baselines with the results obtained by the proposed approach.