The Research And Application Of Data Mining In E-Commerce Recommendation System

Data mining--a kind of fairly new information technology, which integrates knowledge from many subjects such as database? AI and statistics, trying to extract the unknown, effective or useful knowledge from data,has been developed with the technology of database and Artificial Intelligence and emerged as one of the most promising field for database research over the past decade. With the rapid growth and popularization in Internet and WWW, the development of Electronic Commerce has caught more attention from researchers. People expect to make full use of its advantage to gain more economic profit under the new—type commerce. Recommendation system is a successful example of applying data mining technology on the electronic commerce field.By using recommendation system, Electronic Commerce website can analyze the customer's consuming preference, thus recommending proper products to different customers. Recommendation system has become one of the most important technology methods for helping people to get information which customers are really interested in. To meet the personal requirements of consumers, we put forward Collaborative filtering and content-based filtering---the most common information filtering technology in recommendation system. But traditional collaborative filtering algorithm has the shortcomings of sparseness, expansibility, and synonymy. Here, we present a new collaborative filtering algorithm basing on column-vector of the evaluations matrix for an e-book recommendation system which consists of three parts:. Clustering analysis, Collaborative filtering based on wolumn vector and Collaborative filtering based on row vector. First it clusters the goods according to its user characeristics, then reduces dimensions of the user matrix, and finally create recommendation. Not only can it overcome the disadvantages of sparseness, expansibility and synonymy in traditional collaborative filtering algorithm, raise its efficiency performs operation and lower the time, it has also achieved rapid progress in accuracy and validity. Our experiments suggest that the algorithm provides better performance than user-based algorithm. To show its effect on recommendation, the paper just combimes clustering analysis with Collaborative filtering in the book recommendation model system.