Agricultural Meteorological Disaster Relation Extraction Technology Based on BERT and Adaptive Denoising Graph Network
Yonglei Li, Hailong Liu, Qian Xu, Nengfu Xie, Rui Man, Xiaoli Wang, Xin Wang
Preprints.org
Problems Identified (4)
Unstructured disaster text extraction: Rapid growth of unstructured meteorological texts creates challenges for accurate disaster-related information extraction.
GCN noise and redundancy: Relation extraction models integrating semantic and syntactic information with GCNs suffer from noise amplification and information redundancy.
Unstructured disaster text extraction: Rapid growth of unstructured meteorological texts creates challenges for accurate disaster-related information extraction.
GCN noise and redundancy: Relation extraction models integrating semantic and syntactic information with GCNs suffer from noise amplification and information redundancy.
Proposed Solutions (5)
BERT adaptive denoising graph relation extractor: A domain-specific relation extraction framework integrates BERT with an Adaptive Denoising Graph Network.
Dual semantic-syntactic tuning: A dual-tuning architecture jointly models semantic and syntactic features.
SA-GCN dependency pruning: A Self-Attention Graph Convolutional Network dynamically prunes irrelevant dependency structures.
RS-Net adaptive feature denoising: A Residual Shrinkage Network performs fine-grained feature denoising using sample-adaptive threshold learning.
BERT adaptive denoising graph relation extractor: A domain-specific relation extraction framework integrates BERT with an Adaptive Denoising Graph Network.
Results (3)
High F1 on DUIE and disaster dataset:
Outperforms state of the art:
Adaptive denoising ablation effectiveness:
Research Domain
agricultural meteorological disaster relation extraction