Study And Implementation On Prefetch Technique At Proxy Based On Web Mining

The rapid development of WWW changed the method that the information was founded and exchanged. Although the speed of network has been improved considerably in recent years, but the network band-width doesn't come up with the increase of users' demand. Limited by the network band-width, users always suffer from long time waiting. Web prefetching is one of the most popular and effective strategies for reducing the perceived access delay, it has drawn extensive research attention recently.The Web prefetcbing based on Web mining can gaining better performance, because it can show user access behavior more accurate, so it is the focus of Web prefetcbing research continuously. The proxy server, sitting in the middle-tier in the Internet infrastructure, has the effect that reducing the Web latency, if we implement Web prefetcbing based on Web log mining at proxy, it must takes optimization performance.In this thesis, we disusses and summarys web prefetcbing technology based on Web mining at proxy, it's purpose is reducing the web latency and provides web prefetcbing service of individualization for user. In order to improve the performance of Web prefetcbing model, we propose the concept of news Web page according to the characteristic of Web, and remove these news Web pages as noise data when model is created, the experiment indicates that it not only can reduce effectively the size of the models, but can improve the performance of the prediction models as well. In order to trace the incrementally updated of Web, and mend the disadvantage of Web prefetcbing model based on Web log mining that it can't trace the newest change of Web's content and user's behavior, we propose the model of dynamic prediction list and the model of extend dynamic prediction list, they can provide prefetch server for newly increases users according to the change of Web, the experiment indicates that they can farther improve the performance of prediction engine. In order to guarantee the efficiency of Web prefetcbing model, we adopt Hash function and linked list structure to organize model's documents and index.Though the model that we design and implement is based on dependency graph in this thesis, it is suitable for PPM model too, and must gain the better performance.