Browse Books

Reviewer: Claudiu Popescu

User models have an important place in the developing field of human-computer interfaces. Their goal is to create flexible user-oriented dialogue. The user's behavior must be modeled. A further step is to construct online models by interfacing with the user. This collection of papers submitted to an international workshop on user modeling is divided into four parts. Part 1 contains three survey chapters. They provide a good tutorial on user models. The first chapter, by the editors of this volume, deals with the construction, representation, and exploitation of user models. This material is a basic starting point in this domain. The chapter contains the main definitions, some good examples, and references to the main works on each subject. It ends with some unsolved problems and options for future research. The second chapter is by Elaine Rich, who introduced the term user modeling a decade ago. This chapter defines stereotypes: frames with attributes and their values. Rich shows how stereotypes can be a useful tool in creating user models. The third chapter, by Kobsa, deals with the beliefs and goals of users. The author classifies beliefs and goals clearly and mathematically. A table shows how the existing models satisfy each member of this taxonomy. This interesting chapter is followed by a discussion that presents additional criteria for refining the taxonomy. Part 2, “Building User Models,” has four chapters. The papers deal with techniques for implementing relevant facts about the user's beliefs, goals, and plans. The recognition of these facts is important. The first chapter covers KNOME, the user model component of a natural language consultation system for UNIX. The system uses the model to learn how much the user knows about UNIX. A double stereotype system represents the user's expertise and the difficulty of the information required by the user. Much of the chapter deals with deducing the user's level of expertise. The second chapter is about detecting and responding to the misconceptions of the user and has applications in UNIX. The goal is to make the system cooperative so that it becomes an advisor to the user. The author presents an interesting taxonomy of potential explanations for incorrect beliefs of the advisor. The third chapter of this part is about plan recognition. I found it more difficult to read than the other chapters. It presents solutions found in the literature, followed by the author's own research. A special formalism is used for presenting examples. The author gives a strategy for inferring the context model from the dialogue. The fourth chapter is about learning the user's language. Users differ in their linguistic expressions, but each user is consistent over time. This chapter presents a method for generating a specific parser for a particular user. This approach can make the computer friendlier. Three methods of compensating for the limitations of natural language communication are known: (1) requiring the user to learn the interface exactly; (2) extending the interface to various users; and (3) the user-adaptable interface, which meets the user's language. The chapter describes this third method. Examples are from a parser called MULTIPAR. The chapter provides some interesting data about the experience. The examples are described in LISP. Part 3, “Exploiting User Models,” contains four chapters. They examine applications of the theory in more detail. The first chapter gives an example of how encyclopedias are written for adults and for young people. This example is a starting point for a system called TAILOR, which generates descriptions adapted to the user's level of expertise. The second chapter discusses a user model for responding to misconceptions. Examples are taken from interactions with databases and expert systems. The third chapter is a deep analysis of how the user can think about the user model in the interface. It raises some questions and provides some implementation results to answer the questions. The fourth chapter is about educational diagnosis. The last part of the book has the title “Shortcomings of Current Models, Prospects for the Future.” Its purpose is to understand user models better and to build better models. Four chapters make up this part. The first chapter is a good survey of the state of the art; it shows the flaws in current user models. Although the conclusions are pessimistic, the solution is to take a more modest approach to this complex problem. I especially liked the second chapter, which deals with the problem of modeling the user's wants. The authors deal with the last phase of the system development cycle: reorganizing the field of research using the experience that has been gained. They present clear examples and their implementation on the system HAM-ANS. The next chapter deals with student modeling in computer-aided instruction and its relation to user models. The final chapter presents GUMS (a General User Modeling Shell). This user modeling system is intended to provide a set of services for an application system interacting with various users. The system builds and maintains models of the users. This collection of papers is organized as a tutorial. It is addressed to researchers in software, linguistics, and psychology. The field is new and therefore the chapters are rather difficult, but the editors achieve their purpose of presenting an introduction and survey of the area. I was impressed by the bibliography, which provides cross-references between the papers and the pages where they are cited. The subject index is very good: it permits a comparison of how a subject is treated by different authors. The physical form is very good. It is a pleasure to hold such a beautiful book.