The Research Of Hybrid Collaborative Filtering Technologies Based On User-Item

With the wide use of Internet and the speedy development of E-commerce, it brings great convenience to customers. But now customers face the new difficulties caused by the overloading online information.In this situation, customers are hard to search the products they interested and the goods they needed then it will hinder the rapid development of E-commerce. To meet the needs of customers, we research the recommended algorithms of E-commerce recommendation system based on user-centric model. This policy will create greater business value and have a widely applying and studying significance.This paper researches the key technologies of E-commerce recommendation system based on user-centric model—the collaborative filtering technologies in C2B E-commerce. The main job is as follows:(1)Severe challenges on collaborative technologies based on C2B models are analyzed in this paper. It consists of three aspects stand for data sparsity, cold start, extensibility. A new method named"a hybrid collaborative filtering algorithm based on user-item"for solving the problem of data sparsity and cold start is proposed.(2)To improve the user predictive accuracy, new algorithm constructs the statistical rating vector to instead of the traditional user vector and it is more representatives of the customers'characteristic than the existing algorithm; assumes a rating factor S to modify the tranditional similarity measure method, we can get the item similarity matrix which has better accurately ratings of item; fuses the advantages of user-based algorithm and item-based algorithm with the control factorαand it can alleviate the "cold start" problem in the early stages of system.(3) Some corresponding experimental programs are designed to prove the feasibility of the hybrid collaborative filtering algorithm based on user-item. Our used data sets can be downloaded from MovieLens recommendation system. The experimental results show that the improved algorithm can provide better recommendation quality than traditional collaborative filtering algorithms, even with the extreme sparsity of user rating data.