minimizing dialog length in interactive case-based reasoning
David McSherry · International Joint Conference on Artificial Intelligence · 2001
Decision trees induced from stored cases are increasingly used to guide case retrieval in case-based reasoning (CBR) systems for fault diagnosis and product recommendation. In this paper, we refer to such a decision tree as an identification tree when, as often in practice, each of the faults to be identified, or available products, is represented by a single case in the case library. We evaluate common splitting criteria for decision trees in the special case of identification trees. We present simplified versions of those that are most effective in reducing the average path length of an identification tree, or equivalently, the average number of questions asked when the tree is used for problem solving. We also identify conditions in which no such reduction is possible with any splitting criterion.