Oriented E-commerce Personalized Recommendation Technology Research

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 didn'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. Give the customers personalized recommendation about the products is a useful method to address this issue. 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 limitions, for instance, data sparsity and cold start problem.In the view of the deficiency of traditional collaborative filtering technology, this paper has put forward a new personalized recommendation technology combining web log clustering and collaborative filtering. The technology mines users'entire information including implicit interest and explicit interest, so the result of the personalized recommendation can satisfy the users well. First, use cluster algorithm classify the users into different clusters based on users'browse log. Secondly, use a new model of collaborative filtering, which determine the similarity based on users'underlying preferences instead of the surface ratings, it can eliminate the effect of users'different rating schemes. Finally, collaborative filtering based on users'underlying preference using the result of web log clustering, it can solve user data sparsity and cold start problem in some degree, it can make recommendations more accurately. During the course of research in this paper, the main work is as follows: (1)User log clustering. First carry on data preprocessing, it can extract avail information from raw visit logs. Then use cluster algorithm classify the users into different clusters, lay a foundation for searching the nearest neighbor.(2)Collaborative filtering based on users'underlying preference. In order to avoid the problem that two users with similar preference on items may having different rating schemes, the paper converts users'surface ratings into their underlying preferences, competes the similarity of users'underlying preference, makes the result of searching the nearest neighbor more accurately.(3) Personalized recommendation technology combining web log clustering and collaborative filtering. This hybrid recommendation technology may classify the users into different clusters offline, and during collaborative filtering based on users'underlying preference, we can reduce data sparsity efficiently, and improve recommendation quality and speed. This hybrid recommendation technology can provide recommendation service for users whose ratings are fewer, it can solve cold start problem in some degree , makes up the deficiency of traditional recommendation technology.Finally, we designed some relevant experiments and analyzed, the results indicates that our personalized recommendation technology combining web log clustering and collaborative filtering is more advanced than tradition technology in quality of recommendation, it can meet users'demands better.