Explaining CBR Systems Through Retrieval and Similarity Measure Visualizations: A Case Study

Abstract

Explainability in AI is becoming increasingly important as we delegate more safety-critical tasks to intelligent decision support systems. Case-Based Reasoning (CBR) systems are one way to build such systems. Understanding how results are created by a CBR system has become an important task in their development process. In this work, we present how visualizations can help developers and domain experts to evaluate the CBR systems behavior and provide insights to further develop CBR systems in their application scenarios. This paper presents an overview of SupportPrim, a CBR system for the management of musculoskeletal pain complaints, and presents methods that explain its retrieval and similarity measures through visualizations that help to evaluate the system’s performance. In the case study, we conduct experiments within the SupportPrim CBR system using differently weighted global similarity measures to compare their effect on the retrieval. This work shows that providing suitable explanations for the CBR system’s stakeholders increases the likelihood of its adoption, and visualizations allow the creation of different explanations for the different users throughout the development phase, thus allowing for better modeling and usage of the system.