A method for studying the variability of users' thematic profile

This document aims at introducing a method for studying thematic profile variations of Internet surfer. These profiles are characterised by the extraction from a proxy cache of web pages consulted by each user. A semantic analysis of the content of these pages allows us to generate a vector that represents users’ mains interests. This way of processing is close to the Salton representation method [SAL74]. One vector is a sequence of words whose weight reflects the co-occurrence of word sets in pages. These weights are computed and updated as the learning stage analysis of pages progress. Vector is a multidimensional magnitude. To evaluate the semantic closeness between two vectors, we can apply the basic rules of linear algebra such as computing the Euclidean distance the vectors. Some works already published on this subject showed that the level of objectivity of this semantic metric was close to the judgement of real human users [LAN99]. Our goal is to study the temporal evolution of these characterised magnitudes.