The Research Of Personalized Users' Profile Based On Agent And Web Mining

With the fast development of Internet, the on-line information is growing and changing rapidly. Only by web browser we can hardly find the information needed quickly. Search engine is able to resolve the problems to some degree, but it is lack of intelligence and personality. Nowadays another method to resolve the problems is to construct a personalized Web Information Agent System.In this paper the relative work of Web Agent is consulted, and some important factors on the design of these systems are put forward, they are:How to construct the personalized users' profile, what kind of information will be needed, and which Machine Learning algorithm will be selected;The users' interests must be dynamic, how-to adapt to the changing;When the users' profile is formed, how to use it to filter the irrelevant information onusers' interests.Based on these considerations, Author of this paper advances a method, which combines Web Usage Mining with Web Content Mining to establish the personalized uses' profile, and this is a fresh point of view on the construction of user's profile compared with previous work.In this paper, Web Cat, which is a Web Agent system based on client, is set up. In order to catch the users' behavior information in client, we have to develop a new browser, which provides the interfaces to reach this goal.In Web Cat, the Path-Analysis method is used to mine the data of the users'browsing behavior on client. Because of the differences between the client's data and the server's data of the users' browsing behavior, we must modify the Path-Analysis algorithm on server to meet the requirement of the recent application environment. So in our system the Path-Analysis methods include the following steps:1) to identify the browsing process each time; 2)to identify all window sessions in one browsing process; 3) to construct the Extended-Directed-Tree of every . window session to reflect the users' browsing behavior in a window session; 4)to combine all Extended-Directed-Trees to form an Extended-Directed-Forest; 5)to identify "transaction and produce the database of transaction; 6)to mine the frequent path referenced in transactiondatabase, then the users' profile about their browsing behavior is produced.The user's behaviors imply the set of web pages, which is interesting to the user. So the Extended-Directed-Forest, which is transferred from path-analysis module to web-page-learning module, can give us the clue of which web page is a content-page and which is a directory-page based on the set of heuristics in our system. Integrated with the web pages evaluated and collected by the user, we can construct PT , the user's model based on his preferred topic, by using the methods of web content mining.Finally we use the efficiency and accuracy of the recommendation module to prove the effectivity of the users' model. And we proved that our methods were feasible and effective in the set-up of the personalized web users' model.The contributions of this paper are:To advance a method to obtain the users' behavior information on client;To advance a new method to mine the data of the users' browsing behavior on client;.To advance a new method to construct the personalized web users' model combined with users' behavior information and the content of web pages which have been scanned by user;To advance a new method to recommende the hyperlinks according to users' browsing habit and users' preferred topic.