In the early years of any academic discipline, its results are scattered across a wide range of technical reports, conference papers, and papers in diverse journals. During its adolescence, it becomes the focus of conferences, specialized journals appear, and influential papers are gathered together into a book of readings. One of the marks that a discipline has come of age is the appearance of an integrated textbook that abstracts from the contributions of individual researchers to give a coherent account of its principles and methods. This book is such a milestone for the field after which it is named. The author is a prominent researcher in case-based reasoning (CBR), and in this work has taken full advantage of her broad awareness of efforts by other researchers.

CBR is an artificial intelligence technique that solves problems by comparing them with a previously stored library of problems and their corresponding solutions. The technique has been applied to many different types of problems, including diagnosis of and recovery from machine failures, design of industrial schedules, help desk support, legal recommendations, medical diagnosis, and planning. A CBR system stores and indexes known cases (sometimes incrementally, by adding the solutions for new problems as they are derived), compares a presented problem to the case library, identifies known cases that are in some sense similar to the presented problem, and adapts the previous cases to the detailed requirements of the new problem. While there is still much active research in CBR, the technique has matured to the point that it is in widespread industrial use.

The book is organized into five parts, covering a total of 16 chapters. Part 1, “Background,” provides a high-level overview, including a definition of CBR (chapter 1), brief descriptions of half a dozen actual systems (chapter2), a review of different reasoning tasks that CBR can support (chapter 3), and a discussion of the cognitive model behind CBR (chapter 4). The technique is not only a useful mechanism for solving problems but also a serious theory of how people solve problems.

The next three parts discuss different implementation aspects of CBR. Part 2, “The Case Library,” reviews techniques for representing and indexing the solved problems that form the knowledge base of a CBR system. Chapter 5 provides both abstract principles of how cases can be represented and examples from actual systems. A CBR system accesses cases through an index, and the vocabulary used in the index determines what the rest of the system can know about these cases. Chapter 6 discusses how to select an indexing vocabulary, and chapter 7 describes manual, automatic, and hybrid methods for indexing specific cases.

Part 3, “Retrieving Cases from the Case Library,” discusses how to organize and search the case library (chapter8) and how to rank the degree of similarity between a new problem and cases in the library (chapter 9). It also provides further details on indexing as affected by the retrieval problem (chapter 10). Part 4, “Using Cases,” describes the aspects of CBR system operation that take place after the new problem has been matched with cases from the library. Though a retrieved existing case is similar to the new problem, it also differs from it, and chapters 11 and 12 show how it can be adapted to serve as a reliable guide for solving the new problem. Chapter 13 focuses on how cases can be used for interpretation and evaluation. Chapter 14 discusses how goals can be used to direct and manage the reasoning process.

Part 5, “Pulling It All Together,” includes a checklist of issues that a knowledge engineer should take into account in designing and constructing a CBR system (chapter 15), and summarizes the state of the technology and directions for ongoing research (chapter 16).

A detailed appendix describes 83 actual systems, including both research efforts and industrial applications, and will be an invaluable guide for implementers who wish to take advantage of previous experience in their domain. The bibliography includes over 400 references through 1993. The book is clearly written and nicely produced, and will serve as a valuable practitioner’s guide. Instructors willing to provide their own exercises will also find it appropriate as a detailed textbook for students who have already had an introduction to artificial intelligence.