| Presently caching and pre-fetching techniques are the primary solutions used to reduce Web access latency. Web caching technique has been widely used in different places of Internet. But as dynamic documents and personal services increase all over the world, the performance of caching deteriorates significantly. As a result, Web pre-fetching which is an efficient way of making up for Web caching, and the most effective method to break the upper bound of caching performance is becoming a hotspot in Web speedup research area. However there are two problems to be solved before pre-fetching can be put into practice prediction accuracy and pre-fetching control. Prediction mechanism decides the most likely Web objects before the users take the action, pre-fetching control determines which Web objects will be pre-fetched and how many Web objects can be pre-fetched on the basis of the current state of system. Aiming at these two problems, we propose a Structural-Relation-Based Web Pre-fetching model. Therefore, a high prediction precision can be achieved at the cost of relative low storage Complexity and network traffic.First, this thesis introduces the development and the state of the art of Internet and WWW; gives the problems that the Internet faced and corresponding solutions; and the concept, classification and structure of pre-fetching are described; then existing prediction algorithms are summarized. As the emphases the semantic cues of the Web hyperlink are studied, The notional demands of the users' visiting sequence are digged. The behaviors that the users scan the Web represents scan the Web pages basing hyperlinks on the current page, Therefore, we can figure that there is structural-relation among the Web pages; on the other hand, which resemble depth-search of directed graph to some tones, and there is Markov trait when the users skip to scan among web pages. Expanding this behavior of users to a double-random process, which is Hidden Markov Model. That is basic theory of LBSR in the paper.Based on these theories a Web Link-Structural-Relation-Based (LBSR) pre-fetching model is brought forward. LBSR picks-up the semantic cues of hyperlinks contained in the user's requisite information through analyzing the user's visiting sequences, LBSR snatches at a great deal of pages by way of semantic cues from different pages using network crawl software, based on the real character words in common use in the network dictionary. LBSR analyzes the semantic cues of links based on the character words, then the probabilities of the user's visiting sequences are computed, The visiting sequence of the users are analyzed recurring to the understanding technology of nature language. The visiting sequences of the users are changed into the demands of the users' concept. The authority of the current page hyperlinks are computed, LBSR pre-fetches the pages based on the authority.LBSR has two parts: the mining offline subsystem and the pre-fetching online subsystem. The user's visiting sequences are digged by the mining offline subsystem and real time pre-fetching is actualized by the pre-fetching online subsystem. To evaluate the performance of LBSR in the experiment simulations, a few groups of the real Web logs are tested. Exact rate of information picking-up, request hit ratio, session hit ratio, applicable rate, waste rate of bandwidth are defined. The results make it know that the LBSR has a preferably capability, at the same time , it assures pre-fetching veracity. It has preferably usability and high hit ratio and high speed, It can decrease the Web access latency and improve responding speed to some extent. |
