| With the rapid development of Internet technology, the problem of overloading information becomes more and more obviously. How to effectively get useful information becomes an urgent issue. The appearance and application of the personalization system solves this issue, which provides different services to different users. User modeling technology becomes the core of individuation service. The appearance of RSS-based information service changes the way of obtaining information. Furthermore, it can fast provide the newest information to users. In this paper, we mainly focus on researching user modeling and model updating, which based on RSS resources, with our object of improving personalization system's quality.The user model of the article adopts double layer tree-model, its first layer represents as the duality group of user's interest class and user's interest degree, and the second layer user interest subclass uses vector space model to express, which represents as the duality group of user's interests and their weight. By analyzing the relationship between user's behavior and user's interests, we calculate the page interest degree. Then, we can use page's interest degree to update user's interest degree. As time changes, people's interest will keep changing, in order to timely and accurately trace use's interest, we adopt clustering, Rocchio and merge update method to update the user model. According to the model's characteristics, we propose a twice clustering method: combining Hierarchical clustering with K-means clustering. Also, we proposes a method of calculating hierarchical clustering's threshold, in order to improve the speed of clustering. In this paper, we adopt single feedback, times feedback of Rocchio and merge update method to implement user model online and offline learning. Online learning increases system plasticity and offline learning assures system stability.Finally, by designing and implementing a RSS News Filtering system, we test the user model and its update method introduced in this paper. This system makes use of Google's custom search engines to provide much more related information for users successfully. Experimentation data indicates that the method of user modeling and updating is available and effective. |
