Proc. 11th IEEE Intl. Conf. on Tools with Artificial Intelligence, pp. 391-398, Chicago, November 1999
Ontology Based Personalized Search
Alexander Pretschner


Institut fu¨r Informatik
Technische Universita¨t Mu¨nchen
Arcisstraße 21, D-80290 Mu¨nchen, Germany
pretschn@in.tum.de
Susan Gauch
Department of EECS
The University of Kansas
415 Snow Hall, Lawrence, KS 66045, USA
sgauch@eecs.ukans.edu
Abstract
With the exponentially growing amount of information
available on the Internet, the task of retrieving documents
of interest has become increasingly difficult. Search engines
usually return more than 1,500 results per query, yet out of
the top twenty results, only one half turn out to be relevant to
the user. One reason for this is that Web queries are in gen-
eral very short and give an incomplete specification of indi-
vidual users’ information needs. This paper explores ways
of incorporating users’ interests into the search process to
improve the results. The user profiles are structured as a
concept hierarchy of 4,400 nodes. These are populated by
‘watching over a user’s shoulder’ while he is surfing. No ex-
plicit feedback is necessary. The profiles are shown to con-
verge and to reflect the actual interests quite well. One pos-
sible deployment of the profiles is investigated: re-ranking
and filtering search results. Increases in performance are
moderate but noticeable and show that fully automatic cre-
ation of large hierarchical user profiles is possible.
1. Introduction
As of March 1999, the Internet provides about
165 million users worldwide with every imaginable
type of information (source: Nua Internet Surveys,
www.nua.ie/surveys). In general, people have two ways
to find the data they are looking for: they can search, and
they can browse. Search engines index millions of docu-
ments on the Internet and allow users to enter keywords to
retrieve documents that contain these keywords. Browsing

The research presented in this paper was partially supported by the
National Science Foundation CAREER Award number 97-03307. The first
author was in part supported by the German-American Fulbright Program.

This work was carried out while the first author was at the University
of Kansas.
is usually done by clicking through a hierarchy of subjects
until the area of interest has been reached. The correspond-
ing node then provides the user with links to related web-
sites. The search and browsing algorithms are essentially
the same for all users.
It is unlikely that 165 million people are so similar in
their interests that one approach to searching or browsing,
respectively, fits all needs. Indeed, in terms of searching,
about one half of all retrieved documents have been reported
to be irrelevant [3]. The main problem is that there is too
much information available, and that keywords are not al-
ways an appropriate means of locating the information in
which a user is interested. Presumably, information retrieval
will be more effective if individual users’ idiosyncrasies are
taken into account. This way, an effective personalization
system could decide autonomously whether or not a user is
interested in a specific webpage and, in the negative case,
prevent it from being displayed. Or, the system could nav-
igate through the Web on its own and notify the user if it
found a page or site of presumed interest.
This paper studies ways to model a user’s interests and
shows how these models - also called profiles - can be de-
ployed for more effective information retrieval and filtering.
A system is developed that “watches over the shoulder” of
a user while he is surfing the Web. A user profile is created
over time by analyzing surfed pages to identify their content
and by associating that content with the length of the docu-
ment and the time that was spent on it. When pages about
certain subjects are visited again and again, this is taken as
an indication of the user’s interest in that subject. Except
for the act of surfing, no user interaction with this system is
necessary. This paper shows how the profiles can be used to
achieve search performance improvements. The increases
in performance are modest, but they are noticeable, and they
are a first step.
This work has been carried out as part of the OBIWAN
project (www.ittc.ukans.edu/obiwan, [30]) at the University