Study On Personalized Recommendation System For E-Commerce Based On Collaborative Filtering Technique

With the increasingly popularization of Internet and the rapid development of E-commerce(EC), E-commerce provides more and more choices to the customers, but its structure is becoming more and more complex at the same time. E-commerce faces a new challenge: on the one hand, the customer is not very interested in all the products provided by the web site and may browse a lot of pages to find the product he wants; on the other hand, the web site doesn't understand the customer's personalized need and provides the customers the same pages, so it can't enhance the products competitive power and can't maintain the steady relations between the web site and the customer. To address this issue, recommendation systems have been proposed to suggest products and helpful information for consumers. Recommendation systems can enhance E-Commerce sales by converting browsers into buyers, increasing cross-sell and building loyalty to prevent user losing. Recommendation systems have gradually become an important part in E-Commerce. More and more research papers about recommendation systems of E-Commerce appear in many kinds of conferences and journals.The recommendation systems of E-Commerce have been very successful in both research and practice. Collaborative filtering is a successful technology that is implemented in E-commerce recommendation systems today. But when the system scale (such as the structure of the web site, the types of the products or the number of the customers) gradually becomes large, collaborative filtering algorithm has several major limits, for instance, real-time requirement, data sparsity, scalability and cold-start problem.In view of data sparsity of the traditional collaborative filtering, we have analyzed in the reason which collaborative filtering recommendation approach suffered from, then we present an improved collaborative filtering algorithm that combined classification trees and collaborative filtering technology. The main idea of improved algorithm is as follows: First, to divide the item matrix by classification tree algorithm, and to carry on the weighting mapping between the target users and other users separately according to the established classification tree, and to obtain the synthetic score of some items. Using these synthetic score we can calculate the similarity between the target user and other users, and then we take the higher similarity of users as the target user's neighbor. Finally, to predict every item's rating according to the nearest neighbors and to generate recommendation to target user.We design experiment system to test this improved algorithm. The experiment results prove that this algorithm is logical and effective. Compare to traditional collaborative filtering algorithm, this algorithm can overcome the sparsity of user's rating information and generate better recommendation results.