User Clustering Method Based on Personal Preferences to Overcome New User Cold-Start Problem
Sronglong Chhem, Sang-Bong Yoo · 한국지능정보시스템학회 학술대회논문집 · 2014
Collaborative filtering (CF) is one the most popular core methods used in recommender system to provide user a personalized service. However, cold-start problem which happens when new users or items get to the system and do not have any ratings history, presents a serious problem to recommendation system which lead to the lack of accuracy in prediction. In this paper, we propose a novel approach that integrates user preferences extracted from social network with traditional CF to overcome this problem and improve performance of CF. We first cluster user based on genre and production company interests. Then we make predictions by combining the average ratings within the user groups with traditional CF using a coefficient. We conduct an experiment on real dataset in a movie recommendation system and Facebook, and the results show the much improvements of our approach over the traditional CF algorithm in dealing with user cold-start problem.