Abstract:User modeling technology is discussed based on the diversity and randomness of user demands. The key technologies, which include model initialization and update, are researched. The user model is expressed in form of vector space model. It’s built using centroid-based classification method. User’s browsing behaviors are analyzed to get feedback implicitly, calculating the documents degrees. A periodic adaptive learning mechanism based on Rocchio algorithm is put forward. The model’s capability to dynamic tracking user needs is verified using satisfaction as evaluation indicator in experiments.