With the explosive growth of web message, personalized information service becomes a focus for researchers, user interest model is a key technology in personalized service, this paper research tag-based User Modeling contrary to the shortcomings of the current user modeling and the extensive use of tag. Firstly, through some experiments prove that the tags which user using contains stable interest and other characteristics about tags distribution; then presents a weighted tree from coarse to fine granularity describes the user model. To improve the timeliness of service establish weighted frequent itemset table for often used together tags for each user. The model avoids the complex process of extracting key words and expresses user’s interest from the user’s point of view, experiments show that the model can improve the quality of personalized service.