Ramos J, Lipani A. EXtrA-ShaRC: Explainable and Scrutable Reading Comprehension for Conversational Systems. UMAP '24: Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization. 2024; Pages 47-56. https://doi.org/10.1145/3627043.3659546
Summary: This study introduces EXtrA-ShaRC, a dataset designed to enhance the transparency and scrutability of Conversational Machine Reading (CMR) systems. CMR systems interpret context to answer user queries, but often do so in ways that are non-transparent to users. EXtrA-ShaRC addresses this by enabling the system to provide clear explanations for its responses and by allowing users to edit their profiles, with changes reflected in the system's outputs. The authors extend the ShARC dataset and propose a model that simultaneously extracts explanations and answers questions, enabling research into how natural language explanations and counterfactual profiles can improve transparency and user trust in conversational systems.
Comentários