The Research On Self-adaptive Websites Recommendation System Framework Based-on Web Log Mining

As Websites continue growing in size and complexity, a large number of web log records have been collected on WWW server, the results of web log mining have become critical for a number of applications, such as the design of websites system, business and marketing decision support, and the personalization of websites. Self-Adaptive Websites Recommendation System is a application of data mining techniques to Web log data in order to make the Website easy to browse. It can predict the quantity and hobby of future user, and offer the basis of making a serials of wise policy for e-commerce enterprises. In this paper, a Self-Adaptive Websites Recommendation System framework named SAWRS(Self-Adaptive Websites Recommendation System) is presented. It is divided into two parts: the offline part and online part, the former deal with the data collection, pre-process, the frequently accessing mode mining;the latter produces a recommendation collection according to existed mining rule of the offline part and current user accessing behavior, then realizing the adaptive online recommendation services;the main work is as follows:Analyzes data pre-process based on the server log, it is the most essential part in the web mining process, its result directly affects the accuracy and confidence level of the mining algorithm processing results, mainly including below stage: Data Cleaning, User Identification, Session Identificationand and Path Completion, Transaction Identification. This paper realizes the web log data pre-process using the highly effective algorithm of the newest pre-process research results.The Apriori algorithm using iterate searching method by the level, each time iterates produces massive k-item candidate collection, it affected the algorithm efficiency to carry out badly, this article has put forward a optimization method aim at this shortcoming of the algorithm, its core thought is judges its k-1 dimension subset whether existing in k-1 dimension frequent itemset when produced the K-dimension candidate collection, If it existing in it then counts adds 1, otherwise, deletes it directly, if its final counts amount to k then it is the k-dimension frequent itemset. so only need to search k-1 item collection one time when produced a K-item collection every time, it enhanced the algorithm efficiency greatly. The theoretical analysis and the experiment proves this method obviously surpasses the original algorithm.In order to enables the system to automatically realizing self-adaptive recommendation, this paper gain current user accessing path by using the sliding window method, then the utilization association rule algorithm bases on gathers tree's to produce association rule set, after obtains association rule set, the theoretical analysis and experimental result indicated this method is effective and feasible.