Recommendation System Of Commercial Site Research Based On Web Data Mining

With the popularization of network, Electronic Commerce has cought more attention from researchers. They want to take the advantage of the new commerce to gain more customs and profit. But the "data exploding" has become serious. The furious competition under the new commerce requires the real time and deep analysis of all the information data. But how to organize and make use of these information effectively, and how to help the owner of huge data to find out the valuable information and knowledge to guide them to make decision? Web data mining has been combined with Electronic Commerce on this occasion. It is a new branch of data mining and focuses on the research in the Internet on how to find out all implicit knowledge modes among all kinds of data including web logs, user register information, web page etc, and on how to gain some predictive information. Applying web data mining in Electronic Commerce can help a site to improve its service and structure in order to meet the requirement of visitors.It is more difficult to retain old customers and gain new customers in Electronic Commerce, so building good relationship from every aspect of Customer Relationship Management (CRM) and improving customers' loyalty have become more important. Learning user's interest, providing individual service, recommending items that user may needs, and help them to locate their need easily and accurately etc, all can be considered as the representation of sound relation with users. So a commerce site needs a recommendation system to accomplish those functions. Building a recommendation system has got more attention recently, the current paper is just based on the points mentioned above for further expansion in many related fields.First the paper outlines the current research status at home and abroad, introduces the significance of the paper and some relative theories. Then, considering web site's characteristics, user and web data's diversity and distributed computationcharacteristics of network, the paper builds a recommendation system model based on web data mining and relates every function module in detail. Furthermore the paper integrates all modules in recommendation system model into B/S three layers based on CORBA criteria: customer layer, server layer, database layer.Secondly, the paper describes how to gain modes about users' interest and relationship between items through applying web data mining technology which involves web usage mining and web content mining on all web data, such as web server logs, items database, user database, shopping cart etc. In addition, the paper also discusses static modes which are not produced in the way of data mining. They are rule-liked modes and represent the site officers' market strategy thinking. Based on the above-mentioned, the paper puts forward different recommendation strategies which suit both registered users and non-registered users, produces some recommendation formulae on computing a item's recommendation value to a user and related algorithms.The paper gives some validation experiments about the ability of recommendation formulae to depict impact of the objective facts to the system's recommendation. Further there are some programs completed to support the site officers to create static modes in any situations. And last, the paper puts forward the summarizing of this paper and next step's work.