RBRA: A Simple and Efficient Rating-Based Recommender Algorithm to Cope with Sparsity in Recommender Systems

Recommender systems have become an important research area both in industry and academia over the last decade. Memory-based collaborative filtering methods include user-based and item-based methods have been explored in many product domains for their simplicity. Memory-based collaborative filtering methods compute the average ratings between similar users or items to predict unrated entries. As a consequence, it is difficult to find similar users or items when the rating data is sparse. The recommendation quality can be poor. This paper proposed an efficient Rating-Based Recommender Algorithm named RBRA. With a new model based on user behavior and item features, RBRA can achieve results that are more accurate even when the rating data is sparse. In the prediction phase, RBRA takes an adaptively weighted prediction, which utilizes both ratings of the same item by different users and different items by the same user. The final ratings are evaluated from two sources, user-based and item-based approaches. Experimental results show that RBRA achieves 400% faster recommendation speed with better accuracy.