Memory-based collaborative filtering: Impacting of common items on the quality of recommendation

Abstract

In this study, the impact of the common items between a pair of users on the accuracy of memory-based collaborative filtering (CF) is investigated. Although CF systems are a widely used recommender system, data sparsity remains an issue. As a result, the similarity weight between a pair of users with few ratings is almost a fake relationship. In this work, the similarity weight of the traditional similarity methods is determined using exponential functions with various thresholds. These thresholds are used to specify the size of the common items amongst the users. Exponential functions can devalue the similarity weight between a pair of users who has few common items and increase the similarity weight for users who have sufficient co-rated items. Therefore, the pair of users with sufficient co-rated items obtains a stronger relationship than those with few common items. Thus, the significance of this paper is to succinctly test the impacting of common items on the quality of recommendation that creates an understanding for the researchers by discussing the findings presented in this study. The MovieLens datasets are used as benchmark datasets to measure the effect of the ratio of common items on the accuracy. The result verifies the considerable impact exerted by the factor of common items.