Learning User Communities for Improving the Services of Information Providers

In this paper we propose a methodology for organising the users of an information providing system into groups with common interests (communities). The communities are built using unsupervised learning techniques on data collected from the users (user models). We examine a system that filters news on the Internet, according to the interests of the registered users. Each user model contains the user’s interests on the news categories covered by the information providing system. Two learning algorithms are evaluated: COBWEB and ITERATE. Our main concern is whether meaningful communities can be constructed. We specify a metric to decide which news categories are representative for each community. The construction of meaningful communities can be used for improving the structure of the information providing system as well as for suggesting extensions to individual user models. Encouraging results on a large data-set lead us to consider this work as a first step towards a method that can easily be integrated in a variety of information systems.