An Improved Collaborative Filtering Recommendation Algorithm Based on User Forgetting Curve

Abstract

Collaborative filtering recommendation is one of the earliest and successful algorithms in the recommender system. The traditional collaborative filtering algorithm only relies on the user-item scoring matrix for recommendation, and fails to take into account the impact of user interest changes on recommendation weights. An improved similarity based collaborative filtering recommendation algorithm with user interest change is proposed in this paper. The proposed algorithm employed an improved item similarity calculation model and the Ebbinghaus forgetting curve to mimic the user interest tracking. A series of experiments were completed on the Netflix dataset and MovieTweeings dataset, compared with several classic traditional algorithms about the accuracy and coverage rate. The experimental results show that the accuracy of the proposed algorithm overwhelms the traditional algorithm with the improvement of 2%-5%.