A web navigation system based on a neural network user-model trained with only positive web documents

An adaptive system designed to assist in navigating the Web is presented. The core of the system is a user model constructed unobtrusively by observing the user activity and using only positive information to train a certain kind of neural network. The system is built upon neural network techniques designed to attack the problem of user modeling using only positive examples. The system is composed of three main agents: LEARN, CLASSIFY and SHADOW which interact around the neural network model to (respectively) build the user model, apply the user model, and to gather information to train the user model. LEARN has been extensively tested off-WEB on the Reuters data base for information retrieval. CLASSIFY has been used to automaticaily annotate a WEB-browser with recommendations.