Assessing users' interactions for clustering web documents: a pragmatic approach

Abstract

In this paper we are interested in describing Web pages by how users interact within their contents. Thus, an alternate but complementary way of labelling and classifying Web documents is introduced. The proposed methodology is founded on unsupervised learning algorithms, aiming to automatically find natural clusters by means of users' implicit interaction data. Furthermore, it also copes with the dynamic nature and heterogeneity of both users' behaviour and the Web, updating the clustering model over time. We want to show that our framework can be easily integrated in any Website, just employing already-known methods and current technologies.