Problem Sol
ing and Reasoning: Case-based

Malewski A 1964 Generality levels of theories concerning As case-based reasoners process each problem, they
human behavior. In: Zetterberg H L, Lorenz G A (eds.) learn by storing records of their problem solving as
Symposium on Theoretical Sociology. Bedminster, Totowa new cases to guide future reasoning. CBR research
Merton R K 1938 Science, Technology and Society in Se
en-
studies the CBR process both as a model of human
teenth-Century England. Saint Catherine Press, Bruges
Merton R K 1959 Notes on problem finding in sociology. In: cognition and as an approach to building intelligent
Merton R K, Bloom L, Cottrell L S Jr (eds.) Sociology Today: systems. Principles from CBR research serve as a
Problems and Prospects. Basic New York foundation for applied computer systems for tasks
Merton R K 1973 The Sociology of Science. University of such as supporting human decision making, aiding
Chicago Press, Chicago human learning, and facilitating access to electronic
Olson M Jr 1965 The Logic of Collecti
e Action. Harvard information repositories.
University Press, Cambridge, MA
O’Neill J (ed.) 1973 Modes of Indi
idualism and Collecti
ism.
Heinemann Educational, London 1. Case-based Reasoning as Cogniti
e Model
Parsons T 1937 The Structure of Social Action. McGraw-Hill,
New York Initial research on CBR was motivated by obser-
Popper K R 1962 Die Logik der Sozialwissenschaften. KoW lner vations of how humans remember, reason, and learn
Zeitschrift fuW r Soziologie und Sozialpsychologie 14: 233–48. from experiences. People who encounter surprising
[1976 The logic of the social sciences. In: Adorno T W, Albert events are often reminded of similar past events and
H, Dahrendorf R, Habermas J, Pilot H, Popper K R. (eds.) use these remindings to help them understand and
The Positi
ist Dispute in German Sociology. Harper and Row, respond to the new situations (Schank 1982). For
New York] example, a doctor whose patient has an unusual
Schu$ tz A 1933 Der sinnhafte aufbin der sozialen Welt. Springer, reaction to a medicine might be reminded of a previous
Wien [1972 The Phenomenology of the Social World. Heine-
patient who had similar reaction. Remembering the
mann, London]
Samuelson P 1954 The pure theory of public expenditure. Re
iew treatment used for the prior patient suggests a can-
of Economics and Statistics 36: 387–9 didate treatment for the new patient; remembering the
Smith A 1776\1970 The Wealth of Nations. Penguin, Har- outcome of the previous treatment helps to predict
mondsworth; Whitestone, Dublin the new outcome and warns of potential pitfalls to
Trevor-Roper H 1976 Princes and Artists. Thames and Hudson, avoid.
London Psychological studies provide evidence supporting
Ultee W C 1980 Fortschritt und Stagnation in der Soziologie. human use of CBR for a wide range of problem-
Luchterhand, Darmstadt, Germany solving tasks such as learning programming, per-
Ultee W 1997 Do rational choice approaches have problems?
forming mathematical problem solving, explaining
European Sociological Re
iew 12: 167–79
Ultee W 1997 Bringing individuals back into sociology. In: anomalous events, and decision-making (see Leake
Blossfeld H P, Prein G (eds.) Rational Choice Theory and 1998) for examples of this work). Another use of CBR
Large-scale Data Analysis. Westview Press, Boulder, CO is for classification or interpretation. For example, in
Van den Doel H 1979 Democracy and Welfare Economics. legal reasoning in the American legal system, prior
Cambridge University Press, Cambridge cases are the basis for legal judgments: lawyers build
Weber M 1922–3 Gesammelte AufsaW tze zur Religionssoziologie. arguments about how to interpret rules of law by
Mohr, Tu$ bingen [1930 The Protestant Ethic and the Spirit of comparing and contrasting precedent cases to the
Capitalism. Allen and Unwin, London] current situation (see Ashley and Rissland (1988) and
Winckelmann J (ed.) 1978 Max Weber. Die protestantische Ethik
Legal Reasoning Models).
II. Kritiken und Antikritiken. Gu$ tersloh, Mohn, Germany
Ziman J M 1987 The problem of ‘Problem Choice’. Miner
a 25: Research on CBR as a cognitive model studies the
92–106 knowledge and processes that underlie human CBR,
Zuckerman H 1978 Theory choice and problem choice in science. addressing topics such as memory organization, simi-
In: Gaston J (ed.) Sociology of Science, 1st edn. Jossey-Bass, larity assessment, and analogical mapping (Kolodner
San Francisco, CA 1993, Leake 1998). Computational models developed
in this work illuminate functional constraints on
W. C. Ultee reasoning processes and provide testable hypotheses
about human reasoning for psychological investi-
gation.
Problem Solving and Reasoning:
Case-based 2. Steps in Case-based Problem-Sol
ing
Computer CBR systems have been developed for
Case-based reasoning (CBR) is a paradigm of artificial numerous tasks including planning (Hammond 1989,
intelligence and cogniti
e science that models the Veloso 1994), design (Pu and Maher 1997), and
reasoning process as primarily memory based. Case- explanation (Schank et al. 1994). Regardless of the
based reasoners solve new problems by retrieving specific reasoning task, any CBR system retrieves
stored ‘cases’ describing similar prior problem-sol
ing relevant prior cases, adapts their lessons to fit the new
episodes and adapting their solutions to fit new needs. situation, evaluates and applies the result, and retains

12117
Problem Sol
ing and Reasoning: Case-based

the results of processing as new cases to guide future 2.4 Storage

reasoning. Thus, computer models of CBR must
Each processing episode of a CBR system, whether
account for four main processing steps: retrie
al,
resulting in success or failure, is stored to guide future
adaptation, e
aluation\application, and storage.
reasoning. Cases containing successful solutions sug-
gest solutions for similar future problems, while cases
storing failures provide warnings of problems to avoid
2.1 Retrie
al and information on how to avoid them. The storage
In order to facilitate retrieval of relevant cases, stored step packages information about the problem, sol-
cases are organized by indices characterizing circum- ution and its results in a new case. It also assigns
stances in which the cases are likely to be relevant. For indices for the case and places it in the case library
example, in a medical CBR system to suggest treat- (also called the ‘case base’) for future use.
ments for illnesses, treatment cases could be indexed
by the diseases they treat and the characteristics of
patients (e.g., age) that tend to determine whether a 3. Research Issues
given treatment is appropriate. Given a new problem
to solve, a case-based reasoner first performs situation 3.1 Retrie
al and Similarity Assessment
assessment to analyze the input problem and describe
it in terms of the indices used to organize cases. Effective case retrieval depends on appropriate re-
Retrieval algorithms use the description to retrieve the trieval algorithms, well-organized case bases, and
cases in memory expected to be most relevant. For indices that are useful for the current task. In case-
example, nearest-neighbor retrieval algorithms select based problem solving, cases are indexed by infor-
the most similar prior cases according to a metric that mation about the problems they solve. For example, a
weights features by their relative importances. Alter- case-based planning system may index plan cases by
natively, case libraries may be organized into dis- the goals they achieve and constraints they satisfy.
crimination networks to enable retrievals based on Indices should be predictive of the relevance of cases
only a small number of feature comparisons. After and should involve features of problems that are likely
candidate cases have been retrieved, a finer-grained to be known to the reasoner; both abstract and
similarity assessment process is often used to identify concrete features may be useful. A crucial research
the best candidates and to determine differences issue in CBR is developing appropriate indexing
requiring adaptation
ocabularies for particular reasoning tasks. In order to
retrieve useful cases for novel situations, methods are
also needed for reformulating or transforming indices
2.2 Adaptation to widen search when no directly relevant cases are
available. There is also considerable interest in re-
The case adaptation step revises the lessons of the placing traditional measures of similarity with simi-
retrieved case (or revises and combines the lessons of larity criteria that directly reflect pragmatic needs by
multiple retrieved cases) to generate an appropriate predicting which prior cases will be easiest to adapt to
new solution. For example, a doctor might have to fit the new situation (e.g., Smyth and Keane 1998).
adopt a prior treatment by reducing the dosage if the Research on retrieval algorithms investigates methods
prior patient was an adult and the current patient is a for achieving rapid retrievals from large case bases
child, or might combine elements of prior treatments (e.g., by exploiting parallel algorithms).
to address multiple complaints. Computer CBR sys-
tems often use rule-based reasoning to determine how
to transform prior solutions. In interpretive CBR, the 3.2 Case Adaptation
case-based reasoner does not modify the retrieved
case, but instead develops justifications for why the Developing case adaptation rules is often difficult
prior interpretation applies (or does not apply) to the because of knowledge acquisition problems: the
new situation. needed knowledge may not be available or may be
difficult to codify. Consequently, improving case
adaptation methods is a fundamental challenge for
CBR. Research areas include applying machine-
2.3 E
aluation\Application
learning techniques to automatically generate case
Tentative results of case adaptation are evaluated to adaptation rules; applying CBR to a CBR system’s
assess the appropriateness of the new solution and to internal processing, to adapt cases based on prior
determine whether additional changes are needed, adaptation experiences rather than rules; using a
possibly resulting in a cycle of repeated adaptation deri
ational analogy process to account for new cir-
and evaluation. Once a solution is judged acceptable, cumstances by replaying a prior solution-generation
it is applied to the real problem. If the solution fails, it process, rather than revising a prior solution (Veloso
may be repaired, reevaluated, and reapplied. 1994); and applying artificial intelligence methods

12118
 Problem Sol
ing and Reasoning: Case-based

such as constraint-based reasoning to guide the adap- support personnel who diagnose customer problems,
tation process. It is important to note that sufficiently and commercial CBR ‘shells’, are available to facilitate
powerful case adaptation can enable cases to be used the development of such applications (Watson 1997).
in very novel ways; research on flexible retrieval and As novel problems and solutions are added to the case
adaptation processes has given rise to CBR models of base during use, the system builds a sharable corporate
creative problem solving. memory, of cases, aiding corporate knowledge manage-
ment.
CBR is also increasingly applied to electronic
3.3 Case-base Maintenance commerce. For example, case-based support systems
are used during on-line purchases of computers, to
As CBR systems receive long-term use, keeping the help buyers configure their systems by presenting
case base accurate becomes an important problem. similar past configurations as a starting point for them
Likewise, as cases are accumulated, it may be necess- to modify. This helps users find a good match to their
ary to reorganize or update the case base in order to needs, even if they do not know all the important
improve its coverage of important problems and to features of the product they are ordering.
avoid slow retrievals. This has led to considerable In addition, the CBR cognitive model and CBR
interest in methods for improving case quality and principles are guiding the design of educational sys-
retrieval efficiency by strategic deletion, revision, and tems by, for example, suggesting when useful learning
addition of cases to a case base. Because cases are is most likely to occur and how to retrieve examples to
increasingly generated by multiple distributed sources support that learning. Likewise, CBR is being applied
(e.g., different offices of an organization), additional to accessing hypermedia and on-line repositories of
issues concern how to update and access dis- information—providing indexing vocabularies and
tributed case information. example-based query interfaces—and is being used to
focus information retrieval, in order to extract task-
relevant information from large-scale information
3.4 Integration with Other Methods sources. Another active research area, textual CBR,
develops methods for facilitating knowledge acqui-
CBR systems may be integrated with other systems sition by automatically capturing and indexing cases
and approaches—both artificial intelligence systems in textual form. Leake (1996) and Lenz et al. (1998)
and other systems—to leverage off the strengths of present projects illustrating these applications areas.
each. For example, case-based and rule-based reason-
ing may be integrated with each other, with cases used
to guide interpretation of rules or to suggest targets for 5. Relationship to Other Artificial Intelligence
rule-based reasoning. CBR may also be used for speed-
up learning, with initial rule-based problem solving
Methods
building up a case library that can then be used instead Much artificial intelligence research focuses on
of rules, bypassing costly reasoning from scratch, or to methods for generating new solutions from scratch by
avoid errors by storing exceptions to rules. Conversely, chaining together general rules (for example, see
rule-based reasoning may be useful to augment the Expert Systems in Medicine). CBR is an interesting
case library and extend the range of problems that a alternative for a number of reasons, including: (a)
CBR system can solve. simplifying knowledge acquisition (experts often find
it difficult to provide rules for their actions but can
easily provide examples, and CBR systems can be
4. CBR Technology fielded with small case libraries and learn additional
cases during use); (b) providing natural explanations
Applications of CBR include both autonomous and of decisions (by accounting for decisions in terms of
interactive systems. As observed by Kolodner (1993), actual previous examples, rather than in terms of rules
human users are often adept at evaluating the rele- that users may not understand or trust); (c) increasing
vance of prior cases and applying their lessons, but problem-solving efficiency by reusing the results of
may lack the experience needed to acquire a sufficient prior reasoning; and (d) facilitating system develop-
set of cases. On the other hand, case adaptation and ment in domains that are hard to codify in terms of
evaluation are difficult for CBR systems, but practical general rules (e.g., because the rules would have too
methods exist for accessing large case libraries. These many exceptions). In CBR, traditional rule acquisition
complementary strengths have led to increasing focus issues are replaced by issues of capturing and repre-
on case-based ‘retrieve and propose’ systems that senting reasoning episodes, selecting similarity and
support human reasoners by providing them with indexing criteria, and capturing case adaptation
relevant prior cases to aid their decision making (cf. knowledge.
Decision Support Systems). For example, case-based The learning done by CBR systems strikes a middle
‘help desk’ systems are used widely to aid telephone ground between knowledge-poor and knowledge-rich

12119
Problem Sol
ing and Reasoning: Case-based

learning approaches. Unlike knowledge-poor induc- Bibliography

tive learning methods, CBR exploits domain knowl-
Ashley K D, Rissland E 1988 A case-based approach to
edge (e.g., many CBR systems use explanations to modelling legal expertise. IEEE Expert 3(3): 70–7
guide indexing and adaptation). Unlike knowledge- Hammond, K J 1989 Case-based Planning: Viewing Planning as
rich methods, such as explanation-based generali- a Memory Task. Academic Press, Boston, MA
zation, CBR can still be in the absence of a rule-based Kolodner, J 1993 Case-based Reasoning. Kaufmann, San Mateo,
domain theory. CA
In its emphasis on building up knowledge by Leake D B (ed.) 1996 Case-based Reasoning: Experiences,
learning, CBR research is in the same spirit as research Lessons, and Future Directions. AAAI Press, Menlo Park,
on neural networks (see Connectionist Approaches). CA
Leake D 1998 Cognition as Case-based Reasoning. In: Bechtel W,
However, CBR contrasts in storing and reasoning
Graham G (eds.) A Companion to Cogniti
e Science. Black-
from examples rather than storing only learned well, Oxford, UK, pp. 465–76
generalizations. Also, CBR performs ‘lazy learning.’ Lenz M, Bartsch-Spo$ rl B, Burkhard H D, Wess S (eds.) 1998
Rather than immediately generalizing training in- Case-based Reasoning Technology: From Foundations to
stances and discarding the examples, as in, for Applications. Springer, Berlin
example, explanation-based and neural network Pu P, Maher M L (eds.) 1997 Issues and Applications of Case-
learning, CBR stores cases with minimal processing based Reasoning in Design. Erlbaum, Mahwah, NJ
and processes them further only if and when adapta- Riesbeck C K, Schank R C 1989 Inside Case-based Reasoning.
tion is needed to solve new problems. In addition, Erlbaum, Hillsade, NJ
Schank, R C 1982 Dynamic Memory: A Theory of Reminding and
unlike neural networks, case-based reasoning systems
Learning in Computers and People. Cambridge University
can change behavior immediately in response to new Press, Cambridge, UK
examples: a single correct example of dealing with a Schank R C, Riesbeck C K, Kass A (eds.) 1994 Inside Case-
situation, and no further training, is all that is needed based Explanation. Erlbaum, Hillsdale, NJ
for the system to process that situation correctly in the Smyth, B, Keane, M T 1998 Adaptation-guided retrieval: Ques-
future. tioning the similarity assumption in reasoning. Artificial
Intelligence 102(2): 249–93
Veloso M M 1994 Planning and Learning by Analogical
Reasoning. Springer, Berlin
6. Conclusion Watson, I 1997 Applying Case-based Reasoning: Techniques for
Enterprise Systems. Kaufmann, San Francisco, CA
A case-based problem solver solves new problems by
retrieving stored problem cases and adapting their
D. B. Leake
lessons to fit new circumstances; it learns by storing
new episodes for future use. This process can provide
Copyright # 2001 Elsevier Science Ltd.
robust reasoning and learning even in domains that
are hard to codify in rule-based form. CBR research All rights reserved.
has provided computational cognitive models of hu-
man information processing, artificial intelligence Problem Solving and Reasoning,
techniques, and information technology for a wide Psychology of
range of task domains.
The CBR cognitive model is discussed in Kolodner This article discusses two important aspects of human
(1993) and Leake (1998). A volume edited by Leake thinking: how people solve problems and how they use
(1996) surveys the field with chapters on issues and systematic thought to reach conclusions from prem-
applications in tasks such as decision aiding, knowl- ises. The two processes are intermingled closely in
edge navigation, design, diagnosis, and education; most everyday human thinking.
Kolodner (1993) provides an in-depth examination of Woodworth (1938) wrote:
key issues; and a volume edited by Riesbeck and
Schank (1989) includes chapters on important early Two chapters (of 30!) will not be too many for the large topic
CBR research projects and ‘micro-versions’, of the of thinking.
computer programs those projects developed. The
process of developing CBR applications is described Woodworth here understates somewhat the im-
by Watson (1997). A volume edited by Lenz et al. portance he assigned to ‘higher mental processes,’ for
(1998) presents chapters on CBR technology for tasks another chapter examines reading, and nine others,
including supporting electronic commerce and textual memory, and learning. In the second half of the
CBR. twentieth century, the small role of high-level cog-
nition in experimental psychology has been enlarged
See also: Case Study: Logic; Problem Solving and greatly by treating thinking as symbolic information
Reasoning: Case-based; Problem Solving and Reason- processing, formulating theories of thinking as non-
ing, Psychology of; Problem Solving: Deduction, numerical computer programs, and introducing new
Induction, and Analogical Reasoning observational methods and instruments to capture

12120

International Encyclopedia of the Social & Behavioral Sciences ISBN: 0-08-043076-7