Towards Unbiased Evaluation of Detecting Unanswerable Questions in EHRSQL
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917656196546560 |
|---|---|
| author | Yang, Yongjin Kim, Sihyeon Kim, SangMook Lee, Gyubok Yun, Se-Young Choi, Edward |
| author_facet | Yang, Yongjin Kim, Sihyeon Kim, SangMook Lee, Gyubok Yun, Se-Young Choi, Edward |
| contents | Incorporating unanswerable questions into EHR QA systems is crucial for testing the trustworthiness of a system, as providing non-existent responses can mislead doctors in their diagnoses. The EHRSQL dataset stands out as a promising benchmark because it is the only dataset that incorporates unanswerable questions in the EHR QA system alongside practical questions. However, in this work, we identify a data bias in these unanswerable questions; they can often be discerned simply by filtering with specific N-gram patterns. Such biases jeopardize the authenticity and reliability of QA system evaluations. To tackle this problem, we propose a simple debiasing method of adjusting the split between the validation and test sets to neutralize the undue influence of N-gram filtering. By experimenting on the MIMIC-III dataset, we demonstrate both the existing data bias in EHRSQL and the effectiveness of our data split strategy in mitigating this bias. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_01588 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Towards Unbiased Evaluation of Detecting Unanswerable Questions in EHRSQL Yang, Yongjin Kim, Sihyeon Kim, SangMook Lee, Gyubok Yun, Se-Young Choi, Edward Computation and Language Artificial Intelligence Incorporating unanswerable questions into EHR QA systems is crucial for testing the trustworthiness of a system, as providing non-existent responses can mislead doctors in their diagnoses. The EHRSQL dataset stands out as a promising benchmark because it is the only dataset that incorporates unanswerable questions in the EHR QA system alongside practical questions. However, in this work, we identify a data bias in these unanswerable questions; they can often be discerned simply by filtering with specific N-gram patterns. Such biases jeopardize the authenticity and reliability of QA system evaluations. To tackle this problem, we propose a simple debiasing method of adjusting the split between the validation and test sets to neutralize the undue influence of N-gram filtering. By experimenting on the MIMIC-III dataset, we demonstrate both the existing data bias in EHRSQL and the effectiveness of our data split strategy in mitigating this bias. |
| title | Towards Unbiased Evaluation of Detecting Unanswerable Questions in EHRSQL |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2405.01588 |