A Case-Based Model that Integrates Specific and General Knowledge in Reasoning

摘要

This paper introduces ChartD 2, a Congenital Heart Disease Diagnostician that employs a case-based model where specific and general knowledge are combined in reasoning. Specific knowledge is represented in the form of cases while general knowledge is represented in the form of category descriptors. When solving a new case, ChartD 2 uses its general knowledge to draw hypotheses and to guide the search for the most similar cases it has already “seen”. The retrieved cases, representing specific knowledge, are then used to support one of the hypotheses and to justify the conclusion reached. ChartD 2 has been based on an earlier hybrid connectionist/symbolic program called Hycones, developed in the same application domain. Besides enhancing some of Hycones' capabilities, the new system proposes solutions for common problems in Case-Based Reasoning (CBR), such as case matching, indexing and learning. The system ChartD 2 is presented and evaluated, using real cases collected from a medical database. The performance of the system is contrasted with that of Hycones and two other learning algorithms. Moreover, similar research efforts on the use of other sources of knowledge by CBR systems are discussed, and topics for further research are suggested.