David McSherry

Retrieval, reuse, revision and retention in case-based reasoning

Case-based reasoning (CBR) is an approach to problem solving that emphasizes the role of prior experience during future problem solving (i.e., new problems are solved by reusing and if necessary adapting the solutions to similar problems that were solved in the past). It has enjoyed considerable success in a wide variety of problem solving tasks and domains. Following a brief overview of the traditional problem-solving cycle in CBR, we examine the cognitive science foundations of CBR and its relationship to analogical reasoning. We then review a representative selection of CBR research in the past few decades on aspects of retrieval, reuse, revision and retention.

Explanation in Recommender Systems

There is increasing awareness in recommender systems research of the need to make the recommendation process more transparent to users. Following a brief review of existing approaches to explanation in recommender systems, we focus in this paper on a case-based reasoning (CBR) approach to product recommendation that offers important benefits in terms of the ease with which the recommendation process can be explained and the system’s recommendations can be justified. For example, recommendations based on incomplete queries can be justified on the grounds that the user’s preferences with respect to attributes not mentioned in her query cannot affect the outcome. We also show how the relevance of any question the user is asked can be explained in terms of its ability to discriminate between competing cases, thus giving users a unique insight into the recommendation process.

Diversity-Conscious Retrieval

There is growing awareness of the need for recommender systems to offer a more diverse choice of alternatives than is possible by simply retrieving the cases that are most similar to a target query. Recent research has shown that major gains in recommendation diversity can often be achieved at the expense of relatively small reductions in similarity. However, there are many domains in which it may not be acceptable to sacrifice similarity in the interest of diversity. To address this problem, we examine the conditions in which similarity can be increased without loss of diversity and present a new approach to retrieval which is designed to deliver such similarity-preserving increases in diversity when possible. We also present a more widely applicable approach to increasing diversity in which the requirement that similarity is fully preserved is relaxed to allow some loss of similarity, provided it is strictly controlled.

Similarity and compromise

A common cause of retrieval failure in case-based reasoning (CBR) approaches to product recommendation is that the retrieved cases, usually those that are most similar to the target query, are not sufficiently representative of compromises that the user may be prepared to make. We present a new approach to retrieval in which similarity and compromise play complementary roles, thereby increasing the likelihood that one of the retrieved cases will be acceptable to the user. We also show how the approach can be extended to address the requirements of domains in which the user is not just seeking a single item that closely matches her query, but would like to be informed of all items that are likely to be of interest.

Interactive Case-Based Reasoning in Sequential Diagnosis

Interactive trouble-shooting and customer help-desk support, both activities that involve sequential diagnosis, represent the majority of applications of case-based reasoning (CBR). An analysis is presented of the user-interface requirements of intelligent systems for sequential diagnosis. We argue that mixed-initiative dialogue, explanation of reasoning, and sensitivity analysis are essential to meet the needs of experienced as well as novice users. Other issues to be addressed by system designers include relevance and consistency in dialogue, tolerance of missing data, and timely provision of feedback to users. Many of these issues have previously been addressed by the developers of expert systems and the lessons learned may have important implications for CBR. We present a prototype environment for interactive CBR in sequential diagnosis, called CBR Strategist, which is designed to meet the identified requirements.

Retrieval Failure and Recovery in Recommender Systems

In case-based reasoning (CBR) approaches to product recommendation, descriptions of the available products are stored in a case library and retrieved in response to a query representing the user’s requirements. We present an approach to recovery from the retrieval failures that often occur when the user’s requirements are treated as constraints that must be satisfied. Failure to retrieve a matching case triggers a recovery process in which the user is invited to select from a recoveryset of relaxations (or sub-queries) of her query that are guaranteed to succeed. The suggested relaxations are ranked according to a simple measure of recovery cost defined in terms of the importance weights assigned to the query attributes. The recovery set for an unsuccessful query also serves as a guide to continued exploration of the product space when none of the cases initially recommended by the system is acceptable to the user

An Adaptation Heuristic for Case-Based Estimation

Paradoxically, the knowledge acquisition effort associated with rule-based approaches to case adaptation is precisely the overhead that CBR aims to reduce. An adaptation heuristic for case-based estimation is presented which does not rely on domain-specific rules. The approach has been implemented in a case-based reasoner called CREST (Case-based Reasoning for ESTimation) in which the concept of case dominance plays an important role in checking estimates based on the adaptation heuristic and in the maintenance of consistency in the case library. Circumstances in which the adaptation heuristic is appropriate are identified by theoretical analysis and confirmed by experimental results. It is shown to give best results when the value of a case is an additive function of its attributes. The use of domain knowledge to guide the estimation process is examined as a means of enabling the case-based reasoner to cope with departures from this assumption caused by interaction between case attributes.

Advances in conversational case-based reasoning

A considerable amount of research in case-based reasoning (CBR) has recently focused on conversational CBR as a means of providing more effective support for interactive problem solving. We review progress made to date and identify challenges that remain to be addressed.

Incremental relaxation of unsuccessful queries

Increasingly in case-based reasoning (CBR) approaches to product recommendation, some or all of the user’s requirements are treated, at least initially, as constraints that the retrieved cases must satisfy. We present a mixed-initiative approach to recovery from the retrieval failures that occur when there is no case that satisfies all the user’s requirements. The recovery process begins with an explanation of the retrieval failure in which the user’s attention is drawn to combinations of constraints in her query for which there are no matching cases. The user is then guided in the selection of the most useful attribute, and associated constraint, to be eliminated from her query at each stage of an incremental relaxation process. If not prepared to compromise on the attribute suggested for elimination at any stage, the user can select another attribute to be eliminated. On successful completion of the recovery process, the retrieved cases involve only compromises that the user has chosen, in principle, to accept.

Explaining the Pros and Cons of Conclusions in CBR

We begin by examining the limitations of precedent-based explanations of the predicted outcome in case-based reasoning (CBR) approaches to classification and diagnosis. By failing to distinguish between features that support and oppose the predicted outcome, we argue, such explanations are not only less informative than might be expected, but also potentially misleading. To address this issue, we present an evidential approach to explanation in which a key role is played by techniques for the discovery of features that support or oppose the predicted outcome. Often in assessing the evidence provided by a continuous attribute, the problem is where to “draw the line” between values that support and oppose the predicted outcome. Our approach to the selection of such an evidence threshold is based on the weights of evidence provided by values above and below the threshold. Examples used to illustrate our evidential approach to explanation include a prototype CBR system for predicting whether or not a person is over the legal blood alcohol limit for driving based on attributes such as units of alcohol consumed.