Research On User Interest Based On Behavior

It has become an important issue that how to provide people information resource with high quality under the condition of the rapid increase of information quantity. A possible way is to supply the user with the information just of his interest according to his history of browsing information.Obviously, it is a key to acquire and represent the interest of a user appropriately, which includes user information collecting, representation, learning and updating. In this thesis, the collection of user information and the construction and representation of user interest are the main topics.The investigation form was designed and implemented for user interest information collection. When one title of news with a hyper-link in the form was clicked, the internet explorer was activated to open the news accordingly. The time spent on the news and sequences of the news read by the user were recorded automatically. The reader could also issue opinion on the title and the content explicitly.User's behaviors have close relationship with his inner psychology. The user selected and read the news according to his interest. During this process some behaviors of the user is a hint of user interest. Based on this, a reading-time based and a reading-sequence based user modeling approaches were designed and the corresponding algorithms were introduced. And the TF-IDF based and explicit feedback based models were also constructed for performance comparison.A testing system was implemented to construct the above user models. The results of the experiment have shown that reading time was the strongest hint to user's interest, while the reading-sequence reflected the user's interest to some extent. The explicit feedback was not always the real reflection of the interest because of privacy. The TF-IDF based model was better than explicit feedback based one, but worse than the timed-based one. It is clear that reading time is most important, but reading sequence reflects the interest to some extent. When constructing a user's model according to his behaviors, reading time and sequence should be combined together.