E-commerce Recommendation System Based On CBR And Web Log Mining

With the popularization of Internet, E-commerce is developing rapidly and purchasing products from the Internet has become the mainstream. While merchants and customers can enjoy the convenience brought by the network, the phenomenon of information/product overload becomes increasingly serious. Customers are submerged in huge amounts of information since it is not easy for them to find what they need. The merchants also face the problem of losing customers. In this case, the E-commerce recommendation system comes into being.E-commerce recommendation system provides product information and advice for clients through e-commerce websites to help clients to decide what to buy by simulating the behavior of sales staff to help customers to complete the purchase process. The common recommendation techniques include content-based filtering, based on collaborative filtering, based on knowledge discovery, based on utility and so on. Content-based and collaborative filtering techniques are most widely used. But content-based recommendation system, relying on the extraction of characteristics of the project, can only discover the information similar to interests users already have, but can not discover the new resources users may be interested in. Collaborative filtering recommendation system also has some problems such as cold start, sparse evaluation and so on.Analyzing the feasibility to combine CBR (case-based reasoning) and web log mining with E-commerce recommendation system, and integrating the characteristics of CBR and web log mining, we can form a more feasible solution:the easy formal part is handled by the web log mining; Case-based reasoning is used to deal with the description of visualization and those problems difficult to describe structured. Combining web log mining with CBR can make full use of multi-level knowledge and improve the performance of the electronic commerce recommendation system. In addition this paper proposes a framework for the recommendation system adopting B/S structure and design thought of multi-layer. The key technologies of the proposed framework are also given.