Un cadre paraconsistant pour l'{\'e}valuation de similarit{\'e} dans les bases de connaissances

Abstract

This article proposes a paraconsistent framework for evaluating similarity in knowledge bases. Unlike classical approaches, this framework explicitly integrates contradictions, enabling a more robust and interpretable similarity measure. A new measure $ S^* $ is introduced, which penalizes inconsistencies while rewarding shared properties. Paraconsistent super-categories $ \Xi_K^* $ are defined to hierarchically organize knowledge entities. The model also includes a contradiction extractor $ E $ and a repair mechanism, ensuring consistency in the evaluations. Theoretical results guarantee reflexivity, symmetry, and boundedness of $ S^* $. This approach offers a promising solution for managing conflicting knowledge, with perspectives in multi-agent systems.