| With the popularity of Internet and the explosion of WWW, people can ramble through the immense information ocean of the Web. As the result of information explosion, many people with little experience are often lost in the Web, and they feel very distressed because of not finding what they really need. So the puzzle we face now is how to provide better Web personalized recommendation according to people's special need. The aim of Web personalized recommendation is to provide relative pages which people may be interested in when they access Web site. It is the main method of improving service quality and access frequency of Web site.First of all, the thesis sums the Web mining theory, and deeply probes into the concept, categories and studies of personalized recommendation. Besides, it analyses and probes into the typical association rule mining algorithm on which a new algorithm is based. The thesis summarizes the key problems involved in personalized recommendation based on Web mining are as follows: Web mining, personalized recommendation techniques and realizing personalized recommendation based on mining association rule.FP-Mine, a new association rule mining algorithm used in Web personalized recommendation is introduced in this paper, which utilizes FP-Growth's theory and employs the structure of Freq-Set-Tree. The details of the algorithm are described with the help of an example. Meanwhile, the author compares and analyses the difference between FP-Mine and FP-Growth from the point of view of time and space. How to use the mined association rules to achieve Web personalized recommendation and the principles and steps of recommendation prototype are deeply addressed. When the recommended pages are too many, the rule's confidence and lift are taken into account, together with the browsing time spent on recommended pages and the distance between the browsing page and the recommended page. Simultaneously, the weight of the recommended page is counted, and then the recommendation is performed according to the weight of recommended pages. Then, the thesis compares the running time spent on FP-Mine with that of FP-Growth by experiments and confirms that FP-Mine is better than FP-Growth. Finally, the thesis concretely analyses the nature of FP-Mine and concludes that comprehensive measure is the best when the first pages of association rule are from 1 to n by trying it from its coverage, precision and comprehensive survey. |
