A Logical Framework for User-Feedback Dialogues on Hypotheses in Weighted Abduction

Weighted abduction computes hypotheses that explain input observations. It employs parameters, called weights, to output hypotheses suitable for each application. This versatility makes it applicable to plant operation, cybersecurity or discourse analysis. However, the hypotheses selected by an abductive reasoner from among possible hypotheses may be inconsistent with the user’s knowledge such as an operator’s or analyst’s expertise. In order to resolve this inconsistency and generate hypotheses in accordance with the user’s knowledge, this paper proposes two user-feedback dialogue protocols in which the user points out, either positively or negatively, properties of the hypotheses presented by the reasoner, and the reasoner regenerates hypotheses that satisfy the user’s feedback. As a minimum requirement for user-feedback dialogue protocols, we then prove that our protocols necessarily terminate under certain reasonable conditions and achieve a fixed target hypothesis if the user determines the positivity or negativity of each pointed-out property based on whether the target hypothesis has that property.