Research On Multi-Agents Recommender System In E-Commence

After its appearance in 1990s, World Wide Web developed rapidly and amazingly. With the data on Internet increase exponently, it becomes more and more difficult for the user to obtain his exact and useful information. In E-Commerce, for example, from the point of a customer, he needs to select out his favorite items from the overgrowth product databases. While, from the point of the web site, it needs to provide as many as possible items to satisfy different customers with different consuming preferences so as to improve customers' loyalties.This dissertation proposes a model of E-Commerce recommender system based on the multi-agents technology. The main contents are as follow:(1) The paper investigates various main-stream recommender systems and analyzes their strengths and weaknesses. (2) The paper proposes a a model of E-Commerce recommender system based on the multi-agents. Its main idea is that interactive interface agent captures the user's implicit feedbacks and services the user with recommendations, which come from the result set of content-based agent and collaborate agent. With this architecture, most weaknesses could be overcome and advantages could be developed.(3) A research on capturing a user's consuming preferences is conducted. A new model of user profiling and implicit preferences is bring forward as a result, with its implementation. (4) The paper deeply investigates the implicit feedback in personalized service, and divides the implicit feedbacks into three categories: examination, retention and reference. To sum up their effects, an feedback coefficient TFG is generated as an aggregation of various implicit feedback behaviors. (5) In this paper, the functions on item similarity, user similarity and implicit feedback are modified and improved to enhance the recommendation quality. Similarity functions behave with the idea of exacting prominent eigenvalues and filtering out trivial values. (6) As an experimental system, SweetHome supermarket recommender system has been designed by means of the architecture of a hybrid system and algorithms suggested in this paper, along with the experimental data. It is concluded that the multi-agent recommender system could effectively improvepersonalized service in E-Commerce, benefiting both customers and web sites. With greatly confidence, recommender system is a promising and valuable technique for us to engaged in.