Research On Collaborative Filtering In Recommend System Of E-Commerce

Personalized Recommendation has a good application prospect in the area of E-Commerce. It's a very important research field of E-Commerce technologies. Recommender System makes recommendations to user by predicting interest of the user, it helps user to filter information. Collaborative Filtering is frequently used in solving information overload problem, Collaborative Filtering is a main tool used in Personalized Recommendation. The traditional Collaborative Filtering has shortcoming as follows: accuracy, data sparse and cold-start.In this paper, the concept of E-Commerce and Personalized Recommendation in E-Commerce is discussed first, after this, it analyses problems exist in traditional Collaborative Filtering. A method to conquer these problems is proposed based on these analyses.Traditional Collaborative Filtering simply fills a fixed value to build user evaluation matrix. In this paper, the level of items is considered. It builds the user evaluation matrix by calculating the RF/IRF of different levels, where high RF/IRF value will have fixed value. This Method solves the sparse problem and cold-start to a certain extent. The recommend formula for the active user not only concerns the rating of neighbor users but also mixed with mean rating scores of all users. This paper improves the SVD CF Algorithm proposed by Sarwar. Experiments results show the quality of recommendation has been improved in the metrics been used.The algorithms will be used in Jiangxi Xinhua Bookstore Agent-based Personalized Information Service System.