Internet search engines identify web pages that contain user-specified keywords, and then rank these pages according to their (heuristically assessed) relevance to the user’s query. In this paper, we investigated the possibility of evaluating this relevance by the similarity of the returned web page to web pages previously visited by the same user: these previously visited pages thus serve as positive training examples from which a machine-learning program induces an internal model of the user’s interests and preferences. We describe two different ways to represent this model. Our experiments indicate that this approach can indeed improve the ranking.
Wadee S. Alhalabi, Miroslav Kubat, Moiez A. Tapia