期刊名称：Bulletin of the Technical Committee on Data Engineering
出版年度：2018
卷号：41
期号：2
页码：47-58
出版社：IEEE Computer Society
摘要：Explanations are an integral part of human behavior: people provide explanations to justify choicesand actions, and seek explanations to understand the world around them. The need for explanationsextends to technology, as semi-automated and fully-automated systems support crucial activities andincreasingly important societal functions. The interpretability of these systems and the ability to explaintheir decision processes are crucial in developing trust in the systems’ function. Further, explanationsprovide opportunities for systems to interact with human users and obtain feedback, improving theiroperation. Finally, explanations allow domain experts and system developers to debug erroneous systemdecisions, diagnose unexpected outcomes, and improve system function. In this paper, we study andreview existing data integration systems with respect to their ability to derive explanations. We presenta new classification of data integration systems by their explainability and discuss the characteristics ofsystems within these classes. We review the types of explanations derived by the various data integrationsystems within each explainability class. Finally, we present a vision of the desired properties of futuredata integration systems with respect to explanations and discuss the challenges in pursuing this goal..
