| | --- |
| | license: mit |
| | --- |
| | |
| | ## Dataset Details |
| |
|
| | This dataset addresses the following question: "Would the information provided in response to Question B give enough information to sufficiently answer Question A". Thus, this is a directional dataset and match annotations may not hold in reverse. |
| |
|
| | We presented this question to a doctor who determined if the answer to Question B would be sufficient to answer Question A. |
| | In other words, our annotator would imagine they wrote Question A and determine if they would likely obtain everything they need to know from Question B instead. Physicians ask patient questions in very specific ways towards eliciting very specific types of information, making seemingly similar statements non-matches and vice-versa. |
| |
|
| | This dataset was used to help evaluate the methods in our paper "Follow-up Question Generation For Enhanced Patient-Provider Conversations". |
| |
|
| | #### Who are the annotators? |
| |
|
| | This dataset was annotated by a family medicine physician with 20+ years of experience at a large regional hospital in the US. |
| |
|
| | ## Citation |
| |
|
| | If you use this dataset in your work, please cite the following paper: |
| |
|
| | <pre> ``` |
| | @misc{gatto2025followupquestiongenerationenhanced, |
| | title={Follow-up Question Generation For Enhanced Patient-Provider Conversations}, |
| | author={Joseph Gatto and Parker Seegmiller and Timothy Burdick and Inas S. Khayal and Sarah DeLozier and Sarah M. Preum}, |
| | year={2025}, |
| | eprint={2503.17509}, |
| | archivePrefix={arXiv}, |
| | primaryClass={cs.CL}, |
| | url={https://arxiv.org/abs/2503.17509},} |
| | ``` </pre> |
