Lies, damned lies and stereotypes: pragmatic approximations of users Judy Kay ?
Basser Department of Computer Science
University of Sydney
[email protected]

Abstract

Stereotypes are a pervasive element of much work in user modelling. This paper discusses ways that stereotypes have been used, both in user modelling research and, by other names, in manyother areas.
As a basis for better understanding of stereotypes for user modelling, the paper deve lops bases for their uses in such dive rsi?ed areas as information ?ltering, help systems, advisors and the tailors of information presentation. It also deals with major issues for deploying stereotypes: technical issues likethe representation and acquisition of stereotypes and matching individuals to them and also socio-political issues.
The paper describes projects that have dev eloped toolkits for the various technical components of the tasks. These include extensive and diversi?ed approaches to building the stereotypes, determining which to apply to individual users and exploiting them in individualising the user'sinteraction with the machine. Also, mindful of the lies that are inherent in the approximation that a stereotype must be, it discusses tools and approaches for attending to the socio-political concerns.

Introduction

The stereotype is one of the common elements in much user modelling work. It captures default information about groups of people. This simple but powerful idea wasintroduced by Rich (Rich 1979, 1983, 1989) who used people'sdescriptions of themselves to deduce the characteristics of books that theywould probably enjoy. Since Rich'sintroduction of the notion of a stereotype, it has become a basic element in manyuser modelling systems. In particular,manyuser modelling shells support entities that are called stereotypes. Since these are the systems that are designed to use in a range of user modelling applications, we would expect them to represent the major approaches in user modelling.
Forexample, GUMS, a Generalised User Modelling System, (Finin 1989) supported a sophisticated stereotype mechanism. This maintained facts and rules in its stereotypes. It further distinguished between de?nite

?Currently at Dept of Computer Sciences, University of Wisconsin, Madison

and default parts of a stereotype. The former must apply to all users in the class. So theyact as a de?nition for that class (egprogrammer stereotype requires that the person programs). By contrast, the default facts act as initial beliefs.
Similarly,BGP-MS (Kobsa 1990) supports a sophisticated stereotype mechanism. The stereotypes can be constructed with the aid of a graphical tool. This helps the system builder see the relationships between the various stereotypes. The system also checks the consistency of the structures created. It has a rule language for managing stereotypes.
Also Brajnik'sUMT (Brajnik, Guida and Tasso 1990, Brajnik and Tasso 1992) supports a knowledge base of stereotypes which are used as default inferences about the user.Each application has a general stereotype plus more specialised ones for different classes of users. Stereotypes are a basic information source in the um toolkit (Kay 1990) where theyare used for initial default information to model the user when nothing better is available.
Stereotypes are a critical part of Orwant'sDoppelganger (Orwant 1993) where he extends it with the notion of communities which are groups of users with many commonalities. A user may be classi?ed as belonging to several communities and where the user'smodel has no explicit information about some aspect, its value is calculated across the communities the user belongs to. It is rather remarkable that these user modelling shells have very little common. Theytakediffering views of the tasks of user modelling and employdifferent representational approaches. Stereotypes constitute a strong point of commonality.
Giventhe apparent importance of stereotypes, it is useful to re?ne our understanding of what theyare, what theyare not and their relationship to other elements of user modelling. This paper does this, ?rst by characterising the intuitive appeal of stereotypes, then by tightening the de?nitions and analysing different classes of stereotypes. From these, it is possible to identify important issues, both technical and non-technical for the effective deployment of stereotypes.