Mining Web User Interest Based On The Analysis Of User Browser Behavior

The Personalization is very hot and active field not only in the science research but business application, today. However, most of these systems have some drawbacks after analyzing them, such as, the shortcoming of embodiment of personalization; the heavy burden of the system; the ineffective difference between the current interest and the Permanent interest; the bad efficiency of study; and so on. Furthermore, most of the system concentrate on the web mining of web log, but that web log is not integrity, and not well and truly, so that it will lost many useful information of client if the research is applied by methods of web mining of web log only.In this thesis, we bring forward a method to form user interest view by the count of the interest degree of user to web page. In the introduction, the actuality and deficiency are presented, then advanced to the forward; In the second chapter, we found that there was close correlation between user's web browser behavior and user's interest for the web page by psychological theory ,Inner Driven Power; so ,in the third chapter, we described the correlation by linear regresssive equation , then through experimentation we proved that this user interest model was tenable, reasonable and effective, The interest of web user would be influenced along with time going and surrounding changing, which should be captured in the information in order to better personalization .In the fourth chapter, for the sake of final getting user Current Interest View well and truly, the calculation methods of weight is improved for more accurate characterization of web page, and the interest measurement of web page is obtained through the analysis of user browsing behavior, then the Web Page Classification Tree is formed according to the Standard Category Tree. Therefore the commendation for personalization is more efficient by methods of it. In the fifth chapter, it is proved that the recall is up to 75% and the precision is up to 72% through many groups of experimentations.