|
Case Based Reasoning (CBR) is a methodology for "learning from experience". Knowledge in a CBR system is composed of
past examples, or cases, stored in a database or by some other convenient means, known as a casebase. To extract information, a target example
is created, and the database is searched for the closest matches to the target. For example, consider a call center providing
computer technical support. When a customer calls in with a computer problem, the agent fills in a form on her computer screen describing
whatever information is known. The CRM system then searches a database of previous solved problems to find the closest matches and
suggests questions to ask the caller. As more and more information is filled in at each step, the database is repeatedly searched for
closer and closer matches until a solution is found. In this way the accumulated expertise of past problems is applied to the
problem with very little expertise required of the agent.
In effect, the operative principle of CBR is this: Reasoning by remembering.
The Case Based Reasoning Processing Cycle
The Case Based Reasoning process is usually described as 5 steps: Representation, Retrieval, Reuse, Revision, Retention.
The CBR system continuously builds its expertise and adapts as new examples are stored (See Figure).
The Case Based Reasoning Cycle
from A. Aamodt, E. Plaza (1994); Artificial Intelligence Communications, IOS Press, Vol. 7: 1, pp. 39-59.
Representation is a key issue for selecting CBR as the technology implementation choice.
CBR systems are able to represent perhaps the widest range of data types of all the knowledge representation technologies.
While neural networks are limited to numeric data types, and expert systems are limited to numeric and symbolic types, CBR
systems can represent virtually any type of data provided. Some typical data types handled by CBR are numeric, symbolic,
categorical, heirarchical, and text. Any data type that can provide a method of matching one example to another can be effectively
used by a CBR system. Note: some commercial CBR products offer nothing more than text matching.
Retrieval is the process of finding the cases in the casebase that most closely
match the current information known. The current information is represented as a new case with much of the information missing.
This step of matching cases is perhaps the most crucial step in the CBR methodology.
Reuse is the step where matching cases are compared to the new case to form a
suggested solution.
Revision is the testing of the suggested solution to make sure it is suitable and accurate.
The result is a solution that has been confirmed by testing.
Retention is the storage of new cases for future reuse. A strong advantage of CBR over other
reasoning technologies is the fact that new knowledge is continuously and easily added to the store of experience. This step
is usually as simple as storing the new case into the casebase.
Case Based Reasoning is best applied to problems where experts would explain how to solve a problem by giving examples.
It is easiest to apply when a database of past cases exists, but can be "bootstrapped" from only a few examples, as well.
Application areas where CBR has been shown to excel are:
- Call Centers / Help Desks:This is an excellent area of application for CBR. As second tier agents
with advanced expertise add newly solved cases to the database, this expertise is immediately available to less experienced first
tier agents. More problems are solved in the first tier and they are solved more rapidly due to the efficient use of computerized
expertise.
- Planning, scheduling, assembling and designing:For example, a bill of materials and cost estimate
can be generated by matching a new set of specifications to past projects that were similar.
- Diagnosis:A set of symptoms is matched against known examples. Diagnosis includes machine faults and
process deviations, as well as medical applications.
- Legal Reasoning:Legal reasoning is heavily based on precedent. CBR uses past cases to aid in accessing
and displaying these precedents.
To see a demonstration of a deployed Case Based Reasoning System, go to Technology Demonstrations
To learn about other technologies Peer Science employs, select one of the following.
Neural Networks
Fuzzy Logic and Fuzzy Expert Systems
Evolutionary Programming and Genetic Algorithms
| 