Research On Web Usage Mining Based-on E-commerce Log

With the popularization of Intenret and the development of e-commerce, the business of e-commerce are confronted with more and more intense competition. E-commerce site provides abundant data source for data mining, thus more and more merchants are concern on how to make use of data mining technology which carries on the excavation to the web servers log data and the transaction data, from that extraction interest pattern in order to better understand the customer visit behavior, or improves site structure and provides more personalized recommendation service to customers. Therefore, to carry out the research have certain application value and significance.The paper first outlined the web usage mining. Introduced two key technologies of web log mining—sequential pattern mining and cluster analysis, analyzed the advantages and disadvantages of the algorithm, established theoretical principle for the following application.Then, detailed discussed data preprocessing process of web log mining, including data cleaning, user identification, session identification, Frame filter, path supplement and so on.Then, for tradition matrix clustering algorithm carries on the optimization, improved as weight matrix cluster algorithm.This algorithm discretes both user browsing time and user click times , to obtain the weighted visit matrix. Separately to the customer and page's cluster analysis, customer cluster discovers the similar customer community, find latent customer, page cluster classifies the page which content relate. After that, through analysis each kind of user visit log,discovers each kind of user's maximum forward path database.uses GSP algorithm to find each kind of user frequent access path. Finally, applies the clustering and sequence pattern mining's results to e-commerce recommendation system,gives a personalized recommendation system PRS prototype structure.The experiment indicated that the improvement of the weight matrix cluster algorithm has high accuracy and expansibility. Use the mining results apply in personalized recommendation system idea is effective and feasible.