Unifying Used-based and Item-based Algorithms to Improve Collaborative Filtering Accuracy

摘要：

Memory-based collaborative filtering is by calculating for the average ratings of the similar users or items to predict new rating. However, a number of similar ratings from users or items are not available, due to the sparse user-item rating matrix. Consequently, it is very poor to predict ratings accurately. This paper proposes a novel algorithm to overcome the sparsely of user-item rating matrix then calculate prediction rating using the new user-item matrix by combining user-based and item-based approach. First, we just use user-based and itembased methods to compute prediction rating respectively, which apply the adopt cosine similarity to compute users or items similarity. Second, we propose a baseline estimate method to process the extremely sparse data. Besides, a parameter is applied to join baseline estimate, user-based and item-based to predict rating of the blanks of useritem rating matrix. Third, we compute the final prediction rating utilizing the new user-item rating matrix. Finally, experimental results show that our proposed novel memory-based collaborative filtering algorithm can greatly improve the accuracy of prediction rating especially for extremely sparse data.

关键词： Recommendation Systems Collaborative filtering User-based Collaborative Filtering Item-based Collaborative Filtering

作者: WuQin LinXin HeLiang

作者单位: Department of Science and Technology East China Normal University Shanghai, China

会议类型: 国际会议

会议名称: 2010 International Conference on Future Information Technology(2010年未来信息技术国际会议 ICFIT 2010)

会议地点: 长沙

会议语种:英文

页码: 902-907

在线出版日期: 2010-12-14（万方平台首次上网日期，不代表论文的发表时间）