Introduction to the Special Issue on Explanation in Case-Based Reasoning | Leake & McSherry, Artificial Intelligence Review (2005) 24: 103–108
Motivation:
It has long been recognized that examples can play a central role in the effectiveness of explanations provided to people (e.g., Rissland et al., 1984). In fact, one of the attractions of CBR is that the experience captured in cases provides a natural basis for precedent-based explanations of CBR solutions. While rule-based explanations of an intelligent system’s conclusions remain an important legacy from expert systems research, recent research supports the view that an explanation based on experience may be more convincing than an explanation based on rules (Cunningham et al., 2003). However, simply presenting examples to a user is not a panacea. CBR researchers have recently questioned the effectiveness of precedent-based explanations in which the user is simply shown one or more of the most similar cases (Doyle et al., 2004; Maximini et al., 2004; McSherry 2004; Roth-Berghofer 2004). This has led to renewed interest in how to provide the support needed to make better use of the explanatory potential of stored cases. (104)
Topics of papers in the special issues:
- Foundations of explanation in CBR
- CBR systems "use explanations of similarity in the form of symbolic descriptions of similarity and retrieval of the subset of cases that satisfy the description" (105)
All other papers focus on increasing user understanding of a system.
- "Nugent and Cunningham present a CBR approach to explaining the predictions of black box algorithms such as neural networks or support vector machines. Local feature weights derived from a model of the black box algorithm’s behavior in the region of a target problem are used to guide the retrieval of a case that is used to explain the value predicted for the target problem." (3)
- Explanation in mixed-initiative conversational CBR
First paper is Soermo (2005) - Explanation in case-based reasoning .