Three-Way Decision Collaborative Recommendation Algorithm Based on User Reputation

Abstract

Collaborative filtering algorithm is a widely used personalized recommendation technology in e-commerce system. However, due to data sparsity, obtained information is insufficient, so that recommendation accuracy is insufficient. By analyzing user rating data to establish user reputation system, and taking full advantage of user reputation to supplement information contribute to improve recommendation accuracy. In this paper, we use three-way decision to make delayed recommendation and propose an algorithm called Three-way Decision Collaborative Recommendation Algorithm Based on User Reputation (TWDA). Firstly, based on Beta distribution, we introduce three-way decision to the process of calculating user reputation, and we use boundary region parameter to reasonably assign ratings in boundary region into positive or negative region. Then, we combine user reputation with matrix factorization model of collaborative filtering recommendation field. Experimental results on two classic data sets show that TWDA improves recommendation accuracy compared with existing recommendation algorithms.