Event extraction aims to extract structured information from unstructured text. However, when handling complex overlapping and nested events, existing single-stage models face challenges such as task cascading error propagation, long-tail class recognition bottlenecks, and feature representation redundancy. To address these issues, this paper proposed an improved model based on structured cost-sensitive learning and feature redundancy elimination gating. First, in the design of the loss function, the structured focal joint loss was proposed. By introducing a focusing parameter, this mechanism dynamically adjusted the weights of hard samples. We also designed novel task-dependent weighting factors to construct hierarchical gradient constraints, which algorithmically prevented the propagation of upstream trigger detection errors to argument roles. Second, to tackle noise interference during feature fusion, we designed a summation-based gated event fusion layer. This layer utilized Sigmoid gating signals to generate additive interference, replacing traditional linear concatenation strategies to achieve endogenous redundancy elimination while maintaining stable feature dimensionality. Experimental results on the Chinese financial and biomedical datasets show that the F1 score of the proposed model on FewFC improves by 1.6 percentage points compared to the OneEE baseline, outperforming the generative model Degree and the LLM correction strategy LC4EE. On the Genia13 dataset, the argument identification F1 score reaches 80.2%, improvement of 3.4 percentage points. Ablation studies indicate that removing the structured loss leads to a 1.8 percentage points decrease in F1 score. The improved method proposed in this study effectively solves the problems of missing logical constraints and feature noise while retaining the high efficiency of parallel inference, and enhances the accuracy of structured semantic parsing in complex scenarios. However, for extreme long-tail classes with very few training samples, the generalization ability of the model remains to be further enhanced through few-shot learning.
ZHANGBinlong, LEIHaiwei, LIChengqi, et al. Patent entity relation extraction based on the semantics of relation labels and global feature integration[J]. Journal of North University of China (Natural Science Edition), 2025, 46(5): 611-621. (in Chinese)
[4]
GAOJ, YUC, WANGW, et al. Mask-then-fill: A flexible and effective data augmentation framework for event extraction[C]//Findings of the Association for Computational Linguistics: EMNLP 2022, 2022: 4537-4544.
[5]
LIQ, LIJ, SHENGJ, et al. A survey on deep learning event extraction: Approaches and applications[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(5): 6301-6321.
[6]
LIUX, HUANGH, SHIG, et al. Dynamic prefix-tuning for generative template-based event extraction[C]//Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2022: 5216-5228.
WANQizhi, WANChangxuan, HURong, et al. Event extraction based on deep learning: A survey of research issue[J]. Acta Automatica Sinica, 2024, 50(11): 2079-2101. (in Chinese)
HURuijuan, ZHOUHuijuan, LIUHaiyan, et al. Survey on document-level event extraction based on deep learning[J]. Computer Engineering and Applications, 2022, 58(24): 47-60. (in Chinese)
[11]
CHENY, XUL, LIUK, et al. Event extraction via dynamic multi-pooling convolutional neural networks[C]//Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), 2015: 167-176.
[12]
TANEVH, PISKORSKIJ, ATKINSONM. Real-time news event extraction for global crisis monitoring[C]//Natural Language and Information Systems. Berlin, Heidelberg, 2008: 207-218.
[13]
SAKAKIT, MATSUOY, YANAGIHARAT, et al. Real-time event extraction for driving information from social sensors[C]//2012 IEEE International Conference on Cyber Technology in Automation, Control, and Intelligent Systems (CYBER), 2012: 221-226.
[14]
SHENGJ, GUOS, YUB, et al. CasEE: A joint learning framework with cascade decoding for overlapping event extraction[DB/OL]. (2021-07-04) [2025-10-13].
[15]
ZHENGS, CAOW, XUW, et al. Doc2EDAG: An end-to-end document-level framework for Chinese financial event extraction[DB/OL]. (2019-09-23) [2025-10-13].
[16]
CAOH, LIJ, SUF, et al. OneEE: A one-stage framework for fast overlapping and nested event extraction[DB/OL]. (2022-09-06) [2025-10-13].
[17]
HSUI H, HUANGK H, BOSCHEEE, et al. DEGREE: A data-efficient generation-based event extraction model[C]//Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2022: 1890-1908.
[18]
WANGX, LIS, JIH. Code4Struct: Code generation for few-shot event structure prediction[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2023: 3640-3663.
[19]
NGUYENT H, CHOK, GRISHMANR. Joint event extraction via recurrent neural networks[C]//Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2016: 300-309.
[20]
NGUYENT, GRISHMANR. Graph convolutional networks with argument-aware pooling for event detection[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2018, 32(1): 5900-5907.
[21]
CHENY, XUL, LIUK, et al. Event extraction via dynamic multi-pooling convolutional neural networks[C]//Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), 2015: 167-176.
[22]
YANGS, FENGD, QIAOL, et al. Exploring pre-trained language models for event extraction and generation[C]//Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2019: 5284-5294.
[23]
XUN, XIEH, ZHAOD. A novel joint framework for multiple Chinese events extraction[C]//Proceedings of the 19th Chinese National Conference on Computational Linguistics, 2020: 950-961.
[24]
LIUX, HUANGH, SHIG, et al. Dynamic prefix-tuning for generative template-based event extraction[C]//Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2022: 5216-5228.
[25]
GAOJ, YUC, WANGW, et al. Mask-then-fill: A flexible and effective data augmentation framework for event extraction[C]//Findings of the Association for Computational Linguistics: EMNLP 2022, 2022: 4537-4544.
[26]
ZHUM, ZENGK, JIBINGWUJ, et al. LC4EE: LLMs as good corrector for event extraction[C]//Findings of the Association for Computational Linguistics: ACL 2024, 2024: 12028-12038.
[27]
LINT Y, GOYALP, GIRSHICKR, et al. Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision (ICCV), 2017: 2999-3007.
[28]
HENNINGS, BELUCHW, FRASERA, et al. A survey of methods for addressing class imbalance in deep-learning based natural language processing[C]//Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, 2023: 523-540.
[29]
SUZ, LINZ, BAIXUEB, et al. MiLe loss: A new loss for mitigating the bias of learning difficulties in generative language models[C]//Findings of the Association for Computational Linguistics: NAACL 2024, 2024: 250-262.
[30]
MAM D, TAYLORA, WANGW, et al. DICE: Data-efficient clinical event extraction with generative models[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2023: 15898-15917.
[31]
ZHOUY, CHENY, ZHAOJ, et al. What the role is vs. what plays the role: Semi-supervised event argument extraction via dual question answering[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(16): 14638-14646.
[32]
KIMJ D, WANGY, TAKAGIT, et al. Overview of genia event task in bionlp shared task 2011[C]//Proceedings of BioNLP Shared Task 2011 Workshop, 2011: 7-15.