Abstract: Personal information service plays an important role in the Internet, which is based on the user’s interest. The representation and update mechanism for the user profiles are presented in this paper. It collects various types of example documents provided by users and sets up the user profiles based on paragraphs clustered, because the paragraph is the best "grain" to understand the user’s interest and its ability to reveal text topic can be measured by co occurrence in example documents. The documents from stream are matched with the profiles, and then they are delivered to users according to the constraining conditions. Consequently, these relevant documents are used to improve the user profiles.