The role of explanatory relationships in strategies for abduction

We conducted an experiment to test whether explicitly represented knowledge of explanatory relationships can significantly reduce uncertainty and increase correctness in the abductive reasoning process. We compared the performance of four abduction machines, each using a different combination of knowledge types and a different reasoning strategy on an existing knowledge base with three kinds of knowledge: routine-recognition knowledge (precompiled knowledge for pattern-based hypothesis scoring); hypothesis-incompatibility knowledge (two hypotheses cannot both be true); and knowledge of explanatory relationships between hypotheses and data items. We conclude that knowledge-based diagnostic systems can improve their accuracy by explicitly explaining data in addition to the usual pattern-based hypothesis scoring.<
>