自然语言文本中的事件真实性指作者对给定事件存在于客观世界中的确定性程度的描述,正确识别文本中事件的真实性,有助于对自然语言进行深层语义理解。同时,事件真实性检测对诸多自然语言处理应用,如观点检测、事件图谱构建、情感分析等具有重要意义。目前,大多数事件真实性检测研究集中在句子级任务上,而在同一篇章中,经常出现针对同一事件真实性表述不同的情况,此时仅在句子层面识别事件真实性可能会导致矛盾。针对该问题,该文提出了一个基于双层注意力机制的篇章级事件真实性检测方法。首先,利用预训练语言模型BERT对句子进行编码;其次,采用图注意力网络学习句子中的上下文信息与事件之间的依赖关系;最后,利用文档级注意力机制抽取不同句子序列之间的潜在关联,从事件序列集合中获取篇章级事件真实性的最终特征表示。实验结果验证了该方法的有效性,在英文和中文数据集上的实验结果显示,该文所提出方法F1值分别达到87.91%和87.92%,与目前最好系统相比,分别提升了1.40%和1.28%。
Abstract
The event factuality is the certainty degree of a given event described in a text. To address the contradiction in current sentence-level event factuality detection, this paper proposes a document-level event factuality detection approach via a double attention mechanism to deal with the different factual descriptions of the same event. With the sentence encoded by BERT, the graph attention neural network is first applied to learn the dependence between context information and events in sentences. Then, a document-level attention mechanism is adopted to capture the latent correlation features among sequences in the whole discourse. The experimental results on both English and Chinese datasets show the proposed method achieve 87.91% and 87.92% of F1 scores, respectively, which means 1.40% and 1.28% improvements to the state-of-the-art method, respectively.
关键词
篇章级事件真实性 /
图注意力神经网络 /
文档级注意力机制
{{custom_keyword}} /
Key words
document-level event factuality detection /
graph attention neural network /
document-level attention mechanism
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] SAURI R, PUSTEJOVSKY J. FactBank: A corpus annotated with event factuality[J]. Language Resoures and Evaluation, 2009, 43(3): 227-268.
[2] LIU S L, LI Y, ZHANG F, et al. Event detection without triggers[C]//Proceedings of the Annual Conference of the North American Chapter of the Association for Computational Linguistics, 2019: 735-744.
[3] ZHANG J C, QIN Y X, ZHANG Y, et al. Extracting entities and events as a single task using a transition-based neural model[C]//Proceedings of the 28th International Joint Conference on Artificial Intelligence, 2019: 5422-5428.
[4] ZHANG Y J, LI P F, ZHOU G D. Classifying temporal relations between events by deep BiLSTM[C]//Proceeding of the International Conference on Asian Language Processing, 2018: 267-272.
[5] ABOLGHASEM S, GHOLAMREZA M, MAHDY G S, et al. Using syntactic-based kernels for classifying temporal relations[J]. Journal of Computer Science & Technology, 2011, 26(1): 68-80.
[6] LI Z Y, DING X, LIU T. Constructing narrative event evolutionary graph for script event prediction[C]//Proceeding of the 27th International Joint Conference on Artificial Intelligence, 2018: 4201-4207.
[7] GLAVA G, NAJDER J. Construction and evaluation of event graphs[J]. Natural Language Engineering, 2014, 21(4): 1-46.
[8] QIAN Z, LI P F, ZHU Q. Document-level event factuality identification via adversarial neural network[C]//Proceedings of the Annual Conference of the North American Chapter of the Association for Computational Linguistics, 2019: 2799-2809.
[9] HUANG R T, ZOU B W, WANG H L, et al. Event factuality detection in discourse[C]//Proceeding of the Natural Language Processing and Chinese Computing, 2019: 404-414.
[10] DEVLIN J, CHANG M W, LEE K, et al. Bert: Pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of the Annual Conference of the North American Chapter of the Association for Computational Linguistics, 2019: 4171-4186.
[11] VELICKOVIC P, CUCURULL G, CASANOVA A, et al. Graph attention networks[EB/OL].https://arxiv.org/abs/1710.10903[2018-04-04].
[12] SAURI R. A factuality profiler for event factualities in text[D]. Boston: Brandeis University, 2008.
[13] MINARD A L, SPERANZA M, URIZAR R, et al. MEANTIME, the NewsReader multilingual event and time corpus[C]//Proceedings of the 10th International Conference on Language Resources and Evaluation, 2016: 4417-4422.
[14] MINARD A L, SPERANZA M, CASELLI T, et al. The EVALITA 2016 event factuality annotation task (FactA) of the final workshop[EB/OL]. http://ceur-ws.org/Vol-1749/paper_006.pdf[2016-12-7].
[15] QIAN Z, LI P F, ZHANG Y, et al. Event factuality identification via generative adversarial networks with auxiliary classification[C]//Proceeding of the 27th International Joint Conference on Artificial Intelligence, 2018: 4293-4300.
[16] 张刘敏,张赟,李培峰. 基于单句表示的篇章级可信度识别方法[J]. 中文信息学报, 2020, 34(10): 67-75.
[17] ASHISH V, NOAM S, NIKI P, et al. Attention is all you need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, 2017: 6000-6010.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
国家自然科学基金(61703293,61672368,61672367)
{{custom_fund}}
PDF(3015 KB)
