The Research Of User Profile Based On User Behaviors And Genetic Algorithms

With the rapid development of the Internet,the information on it increases as the index number,and the categories of information become more.Researchers have to face the new tasks,such as how to deal with the problem of information overloading and information mazing.Personalized service has now becomed to be a focus research in the domain of information service.To be the core of personalized service,the technology of constructing user profile focuses on how to structure the user's interest source,user profile presentation,user profile updating,user profile storage and so on.In this paper,two new methods of constructing user profile which is based on user behavior and GA is proposed in vector space model.The final experiment validates that the two new methods can provide user profile accurately and effectively.The main research of this paper includes such aspects as below:First,based on analyzing the traditional methods of transfering viewing-time into user interest level ,a new method that transfer Viewing-time into user interest level which is between 0 and 1 nonlinearly based on average time is proposed.It not only fulfil the demand of the method of construcing user profile in this paper,but also is fit for the circumstances that need transfer viewing-time into user interest level.Second, after analyzing the traditional methods that construc user profile with GA, combining the method of transfering viewing-time into user interest level nonlinearly,a new method that construcing user profile based on viewing-time and GA is proposed.It is not only a simple and utility method,but also more efficient and accurate.Third, analyzing other user behaviors and combining linear regression model based on user behavior,a new method that constructs user profile based on user behaviors and GA is proposed.Though its efficience is a little lower, but more accurate than the method based on viewing-time and GA.Finally,the standard classified web pages from the well-known international web site are collected as the testing instances and the performance of the new methods of constructing user profile that are proposed in this paper is compared with each other and a traditional method.The computational results show that the new methods proposed in this paper can construct the user profile accurately and efficiently. The research above has good academic reference value and good applied value in the domain of personalized service.