Proceedings of the Fourteenth Annual Conference of the Cognitive Science Society, 1992, pp. 313{318

Constructive Similarity Assessment:

Using Stored Cases to Define New Situations

David B. Leake
Department of Computer Science
Indiana University
Bloomington, IN 47405
[email protected]

Abstract

A fundamental issue in case-based reasoning is similarity assessment: determining similarities and differences between new and retrieved cases. Many methods have been developed for comparing input case descriptions to the cases already in memory. However, the success of such methods depends on the input case description being sufficiently complete to reflect the important features of the new situation, which is not assured. In case-based explanation of anomalous events during story understanding, the anomaly arises because the current situation is incompletely understood; consequently, similarity assessment based on matches between known current features and old cases is likely to fail because of gaps in the current case's description.

Our solution to the problem of gaps in a new case's description is an approach that we call constructive similarity assessment. Constructive similarity assessment treats similarity assessment not as a simple comparison between fixed new and old cases, but as a process for deciding which types of features should be investigated in the new situation and, if the features are borne out by other knowledge, added to the description of the current case. Constructive similarity assessment does not merely compare new cases to old: using prior cases as its guide, it dynamically carves augmented descriptions of new cases out of memory.

Introduction

Case-based reasoning (CBR) systems facilitate processing of new cases by retrieving stored information about similar prior episodes, and adapting solutions from the prior episodes to fit the new situation (for a selection of current CBR approaches, see (Bareiss, 1991)). A fundamental issue in applying the CBR process is similarity assessment: how to judge the similarity between new cases and those retrieved from memory. The decisions of

whether a retrieved case applies, and of where to adapt it if it fails to apply, depend on similarity judgements; consequently, similarity criteria have been the subject of considerable study. Many approaches have resulted (see (Bareiss & King, 1989) for a sampling), but they share a common property: they compare some subset of the features provided by the input case to features of cases stored in memory.

When input case descriptions contain all the information that is relevant to assessing the applicability of the new case, comparing features in the input case description to the features of old cases works well. However, for the task of case-based explanation construction during story understanding, the input cases presented to the understanding system will seldom provide sufficient information for feature comparisons to determine the relevance of prior cases. Consequently, case-based explanation requires not just comparing a static new case description to stored cases, but elaborating and expanding the new case's incomplete description.

Elaborating the new case requires seeking additional information about the current situation, either by inference from existing system knowledge or by investigation in the world. For example, a detective who knows nothing about person X and is informed of X's death cannot hope to find an appropriate explanation by trying to remember the most similar previous episodes of death|the new case does not yet include sufficient information. Likewise, a story understander facing an anomalous situation is unlikely to begin with explicit knowledge of the important factors to consider during similarity assessment: the central problem for explanation is not matching fixed sets of features, but building up what the new case really is. Thus for both detective and story understander, the information provided by explicit inputs is likely to be too sparse for feature matching to be reliable.