Use Case-Specific Reuse of XAI Strategies: Design and Analysis Through an Evaluation Metrics Library

Nowadays, we have access to a good number of eXplainable Artificial Intelligence libraries and techniques aimed at providing explanations for users to comprehend black-box intelligent systems. However, this presents a double-edged sword: while we can access a wide catalogue of explanation possibilities, determining the most suitable explanation method for a specific situation remains a challenging decision-making task. The iSee project was conceived with the primary goal of constructing a platform where users can share their own experiences with explanations and their successful explanation strategies. Through this CBR platform, other users can leverage these solutions for their own explanation needs, obtaining the most suitable explanation solutions regarding their requirements. In this paper, our focus lies on the reuse step of the CBR cycle. We have developed and implemented constructive reuse approaches, consisting of various methods to assist design users in adapting their solutions to specific use cases and end users. We have validated the applicability of the resulting solutions and introduced an evaluation metrics library designed to assess explanation strategies. Using this library, we have evaluated system solutions based on various key features including computational complexity, popularity, uniformity, diversity, serendipity, and granularity.