A Proposal on the Error Bound of Collaborative Filtering Recommender System

Abstract

We predict accuracy of user's preferences by using memory-based collaborative filtering algorithm in recommender system, and then analyze the results through the EDA approach. The possibilities are presented that prediction accuracy can be evaluated before prediction process by analyzing the results. The classification functions using the generative probability of specific ratings are made, and users are classified by using the classification functions. The prediction accuracies of each classified group are analyzed through statistical tests. The method of setting the Error Bound of users who have high probabilities in low prediction accuracy will be presented.