Affiliations: Dipartimento di Ingegneria dell'Informazione:
Elettronica, Informatica, Telecomunicazioni, University of Pisa, via
Diotisalvi, 2 -56122 Pisa, Italy. E-mail: {p.corsini,f.marcelloni}@iet.unipi.it
Abstract: Determining profiles of web portal typical users can be extremely
useful, for instance, to personalize the web portal, to provide customized
guide and to send tailored advertisements. In this work, we present a system to
produce a small number of user profiles from the web access log and to
associate each user with one of these profiles. The system is based on a
version of the fuzzy C-means (FCM) algorithm which uses the cosine distance
rather than the classical Euclidean distance. After filtering the access log,
for instance, by removing occasional and undecided users, the FCM algorithm
clusters the users into groups characterized by a set of common interests and
represented by a prototype, which defines the profile of the group typical
member. To attest the validity of these profiles, we extract a set of
association rules from the raw access log data by applying the well-known
A-priori algorithm and show how the profiles are a concise representation of
the association rules. Finally, to test the effectiveness of the overall fuzzy
system, we illustrate how the profiles determined by the FCM algorithm from
access log data collected along a period of 30 days allow classifying
approximately 93% of the users defined by access log data collected during
subsequent 30 days.
Keywords: Web mining, user profile, fuzzy c-means, association rules
