An Investigation of Clustering-Based Collaborative Filtering (CBCF)

Abstract

Providing or recommending appropriate content based on the quality of experience is the most important and challenging issues in the system. As collaborative filtering (CF) is the most outstanding and popular technique used in recommender systems, we propose a new cluster-based CF (CBCF) method, which uses only the incentive/punishment user (IPU) model of a given rating it is executed by the user, so it is easy to implement. Our goal is to design a simple cluster-based method. There is no further prior information and the accuracy of the recommendations is improved. To be precise, CBCF with IPU model aims to improve recommendation performance, such as accuracy, recall rate and F1 Score by carefully taking advantage of different preferences between users. Specifically, we set a constraint, we want to maximize the recall (or equal F1 score) optimization problem under given conditions accurate. For this reason, users are divided into several clusters based on actual rating data and Pearson Correlation coefficient. Then, we give rewards/penalties for each item according to our preferences. Trend of users in the same cluster. Our experimental results show excellent performance. An improvement over the benchmark CF scheme, instead of clustering for a given recall rate or F1 score accurate