Gaining Insight through Case-Based Explanation

Abstract

Because CBR is an interpretable process, it is a reasoning mechanism that supports explanation. This can be done explicitly by the system designers incorporating explanation patterns in cases. This can be termed knowledge-intensive explanation in CBR. However, of more interest here is case-based explanation that works by allowing users to consider the relation between different cases. The recommendation of a decision support system can be explained by presenting similar cases that motivate the recommendation. Users can derive insight from similar cases that have different outcomes. The differences in outcome are due to the differences in the un-matching features (provided the effect is not due to noisy data). This is a more knowledge-light approach to case-based explanation. This is appropriate for weak-theory domains where the details of the causal interactions in the domain are not well understood; experts would however be able to express the direction of causal interactions. In this paper we present such a knowledge-light framework for Case-Based Explanation.