Download file (16.5MB)
Download as zip (14.9MB)

コレクション

引用

説明:

We introduce PolyBD-EE, a document-level polymer biodegradation event extraction dataset with structured event annotations and text-grounded participant expansion. We further propose a contradiction-aware evaluation framework that evaluates whether predicted and gold events describe experimentally compatible biodegradation findings under different participant specificity levels.
Using PolyBD-EE, we benchmark multiple modern generative large language models under zero-shot extraction settings. The best-performing setting achieves strict- and lenient-atomic F1 scores of 0.492 and 0.557, indicating that current models can capture underlying biodegradation findings but still struggle with fine-grained participant specificity. Further analysis identifies key bottlenecks, including incorrect participant configurations, over-generalization during participant expansion, and weak separation of event-specific information.

権利情報:

• Creative Commons BY Attribution 4.0 International
  

キーワード: Event extraction, Information extraction, Large language model, Polymer biodegradation, Materials informatics

刊行年月日: 2026-12-31

出版者: Informa UK Limited

掲載誌:

• Science and Technology of Advanced Materials: Methods (ISSN: 27660400) vol. 6 issue. 1 2745582

研究助成金:

• Data Creation and Utilization-Type Material Research and Development Project JPMXP1122714694

原稿種別: 出版者版 (Version of record)

MDR DOI:

公開URL: https://doi.org/10.1080/27660400.2026.2745582

関連資料:

その他の識別子:

連絡先:

更新時刻: 2026-11-05 10:43:18 +0900

MDRでの公開時刻: 2026-11-05 12:28:57 +0900