Improving Dynamic Recommender System Based on Item Clustering for Preference Drifts

Abstract

The recommender system is an efficient tool for online application, which exploits historical user rating on item to make recommendations on items to users. This paper aims to enhance dynamic recommender systems under volatile user preference drifts. It proposed an algorithm to solve sparse data by using Gaussian mixture model to fill in data matrix for sparsity reduction and improve more completely ratings prediction. Subsequently, it utilizes item clustering and linear regression technique to predict the future interests of users in category based and additionally uses the nearest neighbor method to prevent over-fitting. The experimental results show that the proposed approach provides the better performance on rating prediction when compared with the state-of-the-art dynamic recommendation algorithms.