A Model-Based Approach for Supporting Dialogue Inferencing in a Conversational Case-Based Reasoner

Conversational ease-based reasoning (CCBR) a form of interactive case-based reasoning where users input a partial problem description (in text). The CCBR system responds with ranked solution display, which lists the solutions of stored cases whose problem descriptions best match the user’s, and a ranked question display, which lists the unanswered questions in these cases. Users interact with these displays, either refining their problem description by answering selected questions, or selecting a solution to apply. CCBR systems should support dialogue inferencing; they should infer answers to questions that are implied by the problem description. Otherwise, questions will be listed that the user believes they have already answered. The standard approach to dialogue inferencing allows case library designers to insert rules that define implications between the problem description and unanswered questions. However, this approach imposes substantial knowledge engineering requirements. We introduce an alternative approach whereby an intelligent assistant guides the designer in defining a model of their case library, from which implication rules are derived. We detail this approach, its benefits, and explain how it can be supported through an integration with Parka-DB, a fast relational database system. We will evaluate our approach in the context of our CCBR system, named NaCoDAE.