An Agent Model that Intelligently Supports Information Filtering from On-line Newsgroups

摘要

As more and more information becomes available on the internet, users become more and more desirable for tools that will help them deal with the large amount of information and assemble what the users are really interested in, but no need for the users to do a lot. In this paper, we propose an intelligent agent model that supports users for information filtering from electronic network sources. Our agent regularly monitors some information sources on behalf of its user and selects articles based on the user's information preferences learned from the user's feedback. In this agent model, filtering process is divided into two stages: the first stage, which we call initial filtering, does a rough article retrieval just based on checking presence of some keywords in articles; the second stage is called inner learning in which the gathered articles in the first stage are analyzed seriously based on the user's information preferences, which is learned and updated by agent itself based on user's feedback in this stage as well, to select some of interest to present to the user. A Web browser, such as Netscape is selected as the unique user-agent interface in this agent model. In order to serve user's individual information needs well, the agent should be personalized for a user on a specific information topic [9, 15]. A learning algorithm based on a modified TFIDF weighting technique [14, 13, 12] is adopted in our agent model. A case implementation of our agent model which helps user to filter his/her interested conference-related news articles from some USENET newsgroups is also described.