Requirements on Explanations: A Quality Framework for Explainability
Larissa Chazette, Verena Klös, Florian Herzog, Kurt Schneider
Abstract
Explainability has been acknowledged as a fundamental requirement for modern information systems. However, there are currently only few guidelines available to assist software professionals in dealing with this requirement and integrating it into systems. More precisely, there is a lack of frameworks and guidelines that help to define and operationalize explainability requirements. To address this need, we present a quality framework that aggregates external dependencies, characteristics of explanations, and evaluation methods to facilitate the analysis, operationalization, and evaluation of explainability requirements. We conducted a literature study to construct the framework and demonstrated its applicability by using it as a guideline for incorporating explanations into an existing navigation system. Finally, we evaluated the quality and effect of the explanations through an experiment within our case study. Our results show that the quality framework is applicable and beneficial in an industrial context and leads to the construction of explanations that increase usage frequency, system acceptance and user satisfaction.