Research On Personalized Recommendation Based On Web Log Mining

With the rapid development of the Internet applications, there is a sharply increased demand on information services via the Internet. While the huge amount of information distributed on the Internet makes it harder for the individual user acquire what he or she needs. Such a phenomenon could be called information overload. At the same time, the distribution of information on the Internet makes the individual user find useful one more difficult, that is so-called information astray. Being insufficient in positive characteristic and lack for taking users' interest into consideration, most search engines presently have not yet solve problems including information overload and information astray. Until recently, the web personalization service has been proposed so as to settle down those issues.Web log mining is using the data mining technologies mining the Web server log files, we can obtain the knowledge about user access manners, which can be used for Web service providers and Web users. The results can provide kinds of information for improving Web server's design, improving Website server performance, improving personalization service, and so on.The web personalization service based on Web usage mining is the key technique in both researches and applications, mainly using the Web log mining technologies related to different users uses different services strategy, provides different services and personalized service.The dissertation systematic analyses the data preprocessing for Web log mining. With the disadvantage of the user's access interest, this dissertation has proposed a new measurement method based on user's browse time and the pages keywords. This dissertation presents personalized Web recommendation approach based on Interest clustering. According to K-path clustering, more effective path similarity function is given and the interest clustering is based on competitive agglomeration and can determine the best cluster number automatically. A more effective similarity function in recommendation algorithms is given and uses the association rules for providing page recommendation set, designs a personalized recommendation model. The experimental results show that the method can improve the time complexity and can improve the recommendation accuracy and the accuracy is about 87%.The main content in the dissertation is as follows: 1. Data preprocessing method research for Web log mining. This dissertation studies Web log mining data preprocessing process and it's methods, including data cleaning, user identification, session identification, path completion, transaction identification.2. Research the measures and expresses of the user's access interest. This dissertation analyses the present shortage of the style of measures and expresses of the user's access interest. And a new method based on both browsing time and keywords is proposed.3. A new cluster algorithm (CCCA) based on user access interest is proposed. And improves the recommendatory method to overcome the shortage of the present methods. First, according to the shortage of K-paths clustering, more effective path similarity function is given and the interest clustering is based on competitive agglomeration and can determine the best cluster number automatically. The similarity function of recommend algorithm is improved and uses the association rules for providing page recommendation set.4. This dissertation designs a recommendation system model which contains two parts: off-line model and on-line model. It can offer a real-time recommendation service to the users.5. The result of the experimentation shows that the data preprocess algorithms is correct, the interesting clustering algorithms and the personalized recommendation algorithms are better than the algorithms that aren't improved.Finally, this dissertation summarized the author'works and prospected the future farther works.