Case-Based Reasoning, Computational Perspectives

Historical Background

The roots of computational models of CBR date to research currents around the early 1980s, including Roger Schank's studies of reminding and memory, Edwina Rissland's studies of legal reasoning, and Bruce Porter's studies of medical diagnosis. In [Page 129]treating specific stored experiences as the primary knowledge source and the process of adapting prior cases to new needs as the primary reasoning process (rather than the chaining together of general rules), CBR contrasted with rule-based artificial intelligence (AI) approaches. The study of CBR was appealing, both from a cognitive science perspective, for modeling human reasoning, and as a way to further the development of robust and efficient AI systems for reasoning and learning. Reasoning from prior cases may be possible even in poorly understood domains for which it is impossible to account for why the prior case's solution was successful, and CBR may speed up problem solving by reusing prior effort when generating a solution from scratch would be expensive while maintaining the flexibility to adjust for new circumstances.

Types of Case-Based Reasoning

Case-based problem solving characterizes how specific prior solutions are applied to solve new problems. For example, a doctor deciding how to treat a patient with unusual symptoms might be reminded of the successful treatment of a prior patient, a travel planner might develop a new itinerary by adapting an old one, or a labor mediator might start from a prior labor agreement when seeking a new one. Because CBR may use varied criteria to determine which cases to retrieve and to adapt prior solutions, CBR is not limited to straightforward reuse and may lead to creative solutions. For example, an architect may combine aspects of prior designs—or may use information about those designs to select contrasting features—to creatively design a novel building. Case-based models are also applied to interpretation tasks, to classify or analyze new instances by comparing and contrasting them to prior examples. For example, a travel agent assessing whether a client would like a hotel might do so by comparing it to the most similar hotels the client liked or disliked in the past.

The Case-Based Reasoning Cycle

Agnar Aamodt and Enric Plaza characterize the case-based problem solving process as one of retrieval, reuse of an old solution, revision of that solution, and retention of the result. This process forms a closed loop—from retrieving an old case to storing a new case for future retrieval—and is known as the CBR cycle. Each step in the cycle involves research issues, such as how cases are organized in the memory of cases (called the case base), what retrieval algorithms are used, how solutions are adapted to fit new problems, and how the cases are stored.

...