The Research Of Web Personalized Information Recommendation Based On Data Mining

The research of Personalized Active Information Service has made a big progress during these years, the most important of which is personalized information recommendation. Data mining has been applied abroad in recent years as a key research field of artificial intelligence. So, the integration of this two kinds of technology, namely, web personalized information recommendation service based on data mining has becoming an important research task increasingly as the time goes by.In this paper the relative work of the pioneer is consulted, and some important factors on the research of web personalized information recommendation are put forward, they are:· The preprocessing of the logs;· The produce and further treatment of users' profile;· Design the recommendation strategies and recommendation algorithms.The paper discusses preprocessing of web usage mining in detail, and gives a key algorithm of each step. The preprocessing of web usage mining mainly includes: data cleaning, user recognition, session identification , path supplementation and user transaction pattern recognition, etc.This paper designs a personalized information recommendation system based on data mining(WPIRS), and gives a recommendation strategy and algorithms respectively. WPIRS includes: data preprocessing module, mining treatment module, on-line recommendation module and user interface module. We put forward recommendation strategy in WPIRS, and different recommendation algorithmsaccording different users. The system gives two kinds of recommendation algorithms based on association rule mining and user's transaction pattern clustering. The recommendation method based on association rule mining is to take advantage of user's own transaction pattern file to produce assembly tree, then association rules by user's session and assembly tree, finally recommendation sets ensues. It possesses two characteristics: fast speed and exactness. However, it does not fit new users , less-time visitors and users who need fresh information. As to the second kind is to assemble similar user's transaction patterns, produces user's transaction clustering patterns, then match user's session and user's transaction clustering patterns, finally produce recommendation sets. It fits new users, less-time visitors and users who need fresh information.