Mobile Intelligent Agents for Document Classi cation andRetrieval : A Machine Learning

This paper describes an implementation of intelligent, customizable mobile software agents for document classiication and retrieval. The mobile agents are implemented using the Voyager platform. The agents learn user's interests by interacting with the user. Results of experiments using three diier-ent approaches { TFIDF, Bayesian and DistAl (neural network classiier) { for the design of trainable document classiiers are presented. The performance of each classiier with and without feature subset selection (using genetic algorithms) was explored. Experiments with retrieval of journal paper abstracts and news articles demonstrate the feasibility of using machine learning to design mobile intelligent agents for customized information retrieval.