多任务增强的文本生成式事件要素抽取方法

doi:10.3778/j.issn.1002-8331.2401-0076

摘要/Abstract

摘要： 事件要素抽取旨在从非结构化文本中抽取结构化的事件数据，为下游任务提供结构化输入。近年来，许多研究采用预训练语言模型加提示学习的方式实现事件要素抽取，以模板槽位填空的形式完成该任务。然而，以往的研究主要采用单模板单任务的方法，但单一模板难以很好地捕捉事件要素实体间的结构依赖关系，其设计质量会影响最终的抽取结果；并且忽视了在进行多任务学习时，相似任务之间的促进作用。因此，设计了一种基于多任务增强的文本生成式事件要素抽取方法。具体地，分别构建了多模板指令增强任务模块与跨任务协同增强任务模块，以生成式模型作为共享主干网络，多种任务统一训练实现知识高效共享。该方法通过不同模板的指令增强，加强额外语义约束，提高了模型对指令的理解能力，通过跨任务的协同增强，使模型通过不同任务的互相监督，提高了模型对事件文本的理解分析能力。在ACE05数据集和RAMS数据集上的全样本实验中，该方法的Arg-C值分别达到了74.1%和52.4%，达到了最优水平。同时具有优异的少样本性能，在少样本场景下实验，仅需一半的数据量就可以达到阅读理解方法的抽取效果。

关键词: 事件要素抽取, 信息抽取, 提示学习, 多任务学习, 自然语言处理

Abstract: Event argument extraction aims to extract structured event data from unstructured text, serving as a structured input for downstream tasks. In recent years, many studies have adopted pre-trained language models with prompt learning to perform event argument extraction in the form of template slot filling. However, previous research has mainly adopted single-template, single-task methods, which struggle to capture the structural dependencies between event argument entities, and the quality of template design affects the final extraction results. Additionally, these methods overlook the potential benefits of multi-task learning for similar tasks. To address these limitations, this paper proposes a text generation-based event argument extraction method enhanced by multi-task learning. This approach features a multi-template prompt enhancement task module and a cross-task cooperative enhancement task module, using a generative model as the shared backbone network for unified training to achieve efficient knowledge sharing. This method enhances semantic constraints through different template prompts, improving the understanding of instructions of the model. Through cross-task cooperative enhancement, the model benefits from the mutual supervision of different tasks, enhancing its understanding and analysis of event texts. Extensive experiments on ACE05 and RAMS benchmarks demonstrate the effectiveness this approach, achieving great performance on full-samples (74.1% Arg-C score in ACE05 and 52.4% Arg-C score in RAMS) and few-shot settings. In few-shot experiments, only half the data amount is needed to achieve the extraction effect of the MRC method.

Key words: event argument extraction, information extraction, prompt learning, multi-task learning, natural language processing

史张龙, 周喜, 王震, 马博, 杨雅婷. 多任务增强的文本生成式事件要素抽取方法[J]. 计算机工程与应用, 2025, 61(9): 168-176.

SHI Zhanglong, ZHOU Xi, WANG Zhen, MA Bo, YANG Yating. Event Argument Extraction via Multi-Task Enhanced Text Generation[J]. Computer Engineering and Applications, 2025, 61(9): 168-176.

参考文献

[1] AHN D. The stages of event extraction[C]//Proceedings of the Workshop on Annotating and Reasoning about Time and Events, 2006: 1-8.
[2] 胡瑞娟,周会娟,刘海砚,等. 基于深度学习的篇章级事件抽取研究综述[J]. 计算机工程与应用, 2022, 58(24): 47-60.
HU R J, ZHOU H J, LIU H Y, et al. Survey on document-level event extraction based on deep learning[J]. Computer Engineering and Applications, 2022, 58(24): 47-60.
[3] LI Q, LI J, SHENG J, 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.
[4] RAFFEL C, SHAZEER N, ROBERTS A, et al. Exploring the limits of transfer learning with a unified text-to-text transformer[J]. Journal of Machine Learning Research, 2020, 21(1): 5485-5551.
[5] LEWIS M, LIU Y, GOYAL N, et al. BART: denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension[J]. arXiv:1910.13461, 2019.
[6] YANG Y, WU Z L, YANG Y X, et al. A survey of information extraction based on deep learning[J]. Applied Sciences, 2022, 12(19): 9691.
[7] HU R, LIU H, ZHOU H. Role knowledge prompting for document-level event argument extraction[J]. Applied Sciences, 2023, 13(5): 3041.
[8] LI S, JI H, HAN J. Document-level event argument extraction by conditional generation[J]. arXiv:2104.05919, 2021.
[9] HSU I, HUANG K H, BOSCHEE E, et al. DEGREE: a data-efficient generation-based event extraction model[J]. arXiv:2108.12724, 2021.
[10] 贺瑞芳,段绍杨. 基于多任务学习的中文事件抽取联合模型[J]. 软件学报, 2019, 30(4): 1015-1030.
HE R F, DUAN S Y. Joint Chinese event extraction based multi-task learning[J]. Journal of Software, 2019, 30(4): 1015-1030.
[11] CHINCHOR N, MARSH E. MUC-7 information extraction task definition[C]//Proceeding of the Seventh Message Understanding Conference (MUC-7), 1998: 359-367.
[12] DODDINGTON G R, MITCHELL A, PRZYBOCKI M A, et al. The automatic content extraction (ACE) program-tasks, data, and evaluation[C]//Proceedings of the Fourth International Conference on Language Resources and Evaluation (LREC’04) , 2004: 837-840.
[13] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Advances in Neural Information Processing Systems, 2017.
[14] DEVLIN J, CHANG M W, LEE K, et al. BERT: pre-training of deep bidirectional transformers for language understanding[J]. arXiv:1810.04805, 2018.
[15] LIN Y, JI H, HUANG F, et al. A joint neural model for information extraction with global features[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020: 7999-8009.
[16] ZHOU J, ZHANG Q, CHEN Q, et al. A multi-format transfer learning model for event argument extraction via variational information bottleneck[J]. arXiv:2208.13017, 2022.
[17] LIU P, YUAN W, FU J, et al. Pre-train, prompt, and predict: a systematic survey of prompting methods in natural language processing[J]. ACM Computing Surveys, 2023, 55(9): 1-35.
[18] DU X, CARDIE C. Event extraction by answering (almost) natural questions[J]. arXiv:2004.13625, 2020.
[19] 吴旭, 卞文强, 颉夏青,等. 机器阅读理解式中文事件抽取方法[J]. 计算机工程与应用, 2023, 59(16): 93-100.
WU X, BIAN W Q, XIE X Q, et al. Chinese event extraction by machine reading comprehension[J]. Computer Engineering and Applications, 2023, 59(16): 93-100.
[20] SI J, PENG X, LI C, et al. Generating disentangled arguments with prompts: a simple event extraction framework that works[C]//Proceedings of the 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022: 6342-6346.
[21] WANG X, ZHOU W, ZU C, et al. InstructUIE: multi-task instruction tuning for unified information extraction[J]. arXiv:2304.08085, 2023.
[22] HUANG K H, HSU I, NATARAJAN P, et al. Multilingual generative language models for zero-shot cross-lingual event argument extraction[J]. arXiv: 2203.08308, 2022.
[23] 李鸿鹏, 马博, 杨雅婷,等. 基于槽位语义增强提示学习的篇章级事件抽取方法[J]. 计算机工程, 2023, 49(9): 23-31.
LI H P, MA B, YANG Y T, et al. Document-level event extraction method based on slot semantic enhanced prompt learning[J]. Computer Engineering, 2023, 49(9): 23-31.
[24] EBNER S, XIA P, CULKIN R, et al. Multi-sentence argument linking[J]. arXiv:1911.03766, 2019.
[25] MA Y, WANG Z, CAO Y, et al. Prompt for extraction? PAIE: prompting argument interaction for event argument extraction[J]. arXiv:2202.12109, 2022.
[26] HUANG K H, HSU I, PAREKH T, et al. A reevaluation of event extraction: past, present, and future challenges[J]. arXiv:2311.09562, 2023.
[27] LOSHCHILOV I, HUTTER F. Fixing weight decay regularization in adam [J]. arXiv:1711.05101， 2017.
[28] LIU J, CHEN Y, XU J. Machine reading comprehension as data augmentation: a case study on implicit event argument extraction[C]//Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021: 2716-2725.