Failure-Driven Transformational Case Reuse of Explanation Strategies in CloodCBR

Ikechukwu Nkisi-Orji, Chamath Palihawadana, Nirmalie Wiratunga, Anjana Wijekoon, David Corsar

Abstract: In this paper, we propose a novel approach to improve problem-solving efficiency through the reuse of case solutions. Specifically, we introduce the concept of failure-driven transformational case reuse of explanation strategies, which involves transforming suboptimal solutions using relevant components from nearest neighbours in sparse case bases. To represent these explanation strategies, we use behaviour trees and demonstrate their usefulness in solving similar problems. Our approach uses failures as a starting point for generating new solutions, analysing the causes and contributing factors to the failure. From this analysis, new solutions are generated through a nearest neighbour-based transformation of previous solutions, resulting in solutions that address the failure. We compare different approaches for reusing solutions of the nearest neighbours and empirically evaluate whether the transformed solutions meet the required explanation intents. Our proposed approach has the potential to significantly improve problem-solving efficiency in sparse case bases with complex case solutions.