Learning Text Filtering Preferences

Anandeep S. Pannu and Katia Sycara The Robotics Institute School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 [email protected], [email protected] Abstract We describe a reusable agent that learns a model of the user's research interests for ltering conference announcements and request for proposals (RFPs) from the Web. For this task, there is a large volume of irrelevant documents and the proportion of relevant documents is very small. It is also critical that the agent not misclassify relevant documents. Information Retrieval and Neural Network techniques were utilized to learn the model of user's preferences. Learning was boot-strapped using papers and proposals the user had written as positive examples. The agent's performance at startup is quite high. Information retrieval and Neural Nets were used to train the agent and experimental performance results were obtained and reported. Introduction We describe a reusable Learning Personal Agent (LPA) that learns a user's research interests and lters Webbased conference announcements and RFPs. Learning Personal Agents have been used for information ltering from the WWW (Lang 1995), (Armstrong et al. 1995), (Pazzani, Nguyen, & Mantik 1995). In contrast to the environments of these systems where the probability of nding relevant links or documents is relatively high, so that the agent's task is to avoid user information overload, in our environment the probability of nding relevant documents is very small and the cost of missing a relevant document is very high. Our goal is to put in the hands of CMU faculty an LPA to retrieve conference announcements and requests for proposals based on their research interests. Each LPA uses the same learning methods and interface but learns a di erent user model. We have experimented (see Section 6) with the LPA of one user and we are currently setting up the LPA's of other users. The News Information agent (Figure 1) polls the Commerce Business Daily (CBD) and conference announcement newsgroups for arrival of new items. When a new item arrives the News agent forwards it This research was supported by ARPA contract F33615-93-1-1330 Usenet Newsgroup Server