My Visits at Google and IBM Almaden Research

August 16 2007–
This summer I was in the Bay Area. I took the occasion and visit IBM Almaden Research and Google.
I gave a presentation both at IBM and Google on my research project, Gugubarra.

Here are some details:

Title of the talk:
The Gugubarra Project: Building and Evaluating User Profiles for
Visitors of Web Sites.

Speaker: Roberto V. Zicari
Computer Science Department,
Goethe University of Frankfurt, Germany.

Work done in cooperation with Natascha Hoebel, Sascha Kaufmann,
Karsten Tolle, Naveed Mustaq.

Index Terms: User Profiles, Web Communities, User modeling, Clustering methods

—————————————————-
Abstract

In this talk I will present an overview of the work we are
currently doing in the Gugubarra project. The project aims at
building tools for better managing communities of Web visitors.

The Gugubarra project (Gugubarra is the Aboriginal name for the
Kookaburra bird) began in 2004 within the database group (DBIS)
at the Computer Science Institute of the Johann Wolfgang Goethe
University, with the aim to build tools for better managing
communities of registered Web visitors.

In this talk I will present the results of the project so far
and outline some open research issues.

The starting point of our project is the assumption that a
community of users is registered on a Web site and that for each
user a profile is built. A User profile is based on the actions
and navigations the user performs on the Web site.

In Gugubarra, we offer various settings that can be used to
create and manage user profiles. By using these settings the Web
site owner can focus on those aspects he wants to analyze.

The approach we have in Gugubarra is as follows:

i) For each registered Web visitor we create a profile.
These user profiles reflect the “inferred” interests of the
users related to a set of pre-defined topics defined by the owner of
the Web site. The profiles go beyond collecting the obvious
information the user is willing to give at the time of
registration. In Gugubarra, a user profile contains two parts:
the obvious profile, given directly by the user and a non
obvious profile (NOP), inferred by the user’s behavior during his visits
on the site.

ii) A user profile is (re)-calculated dynamically any time an
explicit feedback is given by the user and/or a set of events
occurred which are related to the user’s behavior and to certain
“locations” of the Web site.

iii) We cluster Web visitors by clustering similar profiles of
interest [5]. Cluster of Web visitors can then be used to
analyze patterns of interests in the Web community and to forecast
further behavior. Clusters might also provide useful information
to support the decision what kind of new E-services to introduce
for the Web community and when to introduce them.

A first research prototype system, called Gugubarra 1.0 has been
implemented in 2004 [1], which allows to build and manipulate
non-obvious user profiles. It was showcased at the CeBIT Trade
Fairs in 2004 and 2005. Gugubarra 1.0 works as a test
application on real data provided by the Web community viewzone.org.

A new prototype system, Gugubarra 2.0, is currently being
designed [3], [2] which includes a more sophisticated approach
to the definition of non-obvious user profiles and allows
clustering users by interests [4]. In this talk I will focus mainly on
the new features introduced in Gugubarra 2.0 and refer to [6] for
the features implemented in Gugubarra 1.0
—————————————————–

Resources

[1] Building and Evaluating Non-Obvious User Profiles for
Visitors of Web Sites. N. Mushtaq, K. Tolle, P. Werner and R.
V. Zicari.
IEEE Conference on E-Commerce Technology (CEC 04) July 6-9,
2004,San Diego, California, USA (.pdf 265KB)

[2] The Gugubarra Project: Building and Evaluating User Profiles
for Visitors of Web Sites. N. Hoebel, S. Kaufmann, K. Tolle, R.
V. Zicari.
First IEEE Workshop on Hot Topics in Web Systems and
Technologies (HotWeb 2006), November 13-14, 2006, Boston,
Massachusetts, USA (.pdf 415)

[3] The Design of Gugubarra 2.0: A Tool for Building and
Managing Profiles of Web Users. N. Hoebel, S. Kaufmann, K. Tolle, R. V.Zicari.
IEEE/WIC/ACM, International Conference on Web Intelligence,
2006, Hong Kong (.pdf 100) Long Version: (.pdf 196)

[4] On Clustering Visitors of a Web Site by Behavior and
Interests.
N. Hoebel, R. V. Zicari. in Studies in Computational
Intelligence Series, Springer, AWIC ’07 (.pdf 155)

The above papers can be downloaded at:
Gugubarra Papers

If you are interested you can also view the video of my presentation at Google.