Accumulative Influence Weight Collaborative Filtering Recommendation Approach

Memory-based collaborative filtering algorithms are widely used in practice. But most existing approaches suffer from a conflict between prediction quality and scalability. In this paper, we try to resolve this conflict by simulating the ”word-of-mouth” recommendation in a new way. We introduce a new metric named influence weight to filter neighbors and weight their opinions. The influence weights, which quantify the credibility of each neighbor to the active user, form accumulatively in the process of the active user gradually provides new ratings. Therefore, when recommendations are requested, the recommender systems only need to select the neighbors according to these ready influence weights and synthesize their opinions. Consequently, the scalability will be significantly improved without loss of prediction quality. We design a novel algorithm to implement this method. Empirical results confirm that our algorithm achieves significant progress in both aspects of accuracy and scalability simultaneously.