AFFPD-ONEE: Overlapping and Nested Event Extraction Based on Adaptive Fusion Filtering and Position Decoding
Event extraction is a highly significant information extraction task in natural language processing. Most existing work focuses on flat event extraction, which often fails to effectively address overlapping and nested event extraction, In particular, the role of symmetrical attention between sentenc...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2024-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10792882/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Event extraction is a highly significant information extraction task in natural language processing. Most existing work focuses on flat event extraction, which often fails to effectively address overlapping and nested event extraction, In particular, the role of symmetrical attention between sentences and event types, and the positional relationships between arguments, are often overlooked. To tackle these issues, this paper proposes an overlapping and nested event extraction model, AFFPD-ONEE, based on adaptive fusion filtering and position decoding. Firstly, we propose a trigger extraction method based on adaptive fusion filtering. By leveraging symmetrical bidirectional attention between event types and text, this method enhances the representation of event types and sharpens the focus on key information by filtering nodes, while discarding irrelevant details. This approach enables the model to effectively handle overlapping and nested events with multiple labels, improving its overall generalization. Secondly, we propose an argument extraction method based on a position decoding algorithm. By incorporating position decoding into the extraction process, we enhance the precision of the AFFPD-ONEE model in identifying argument boundaries. This improvement enables the model to more effectively capture relationships within overlapping and nested events. Experiments on public datasets demonstrate that the AFFPD-ONEE model significantly outperforms state-of-the-art (SOTA) models in extracting overlapping and nested events. Our code, data and trained models are available at <uri>https://github.com/18118019819/AFFPD</uri>. |
|---|---|
| ISSN: | 2169-3536 |