Elamur Research⌕
☰
home / papers / 2019 / Doc2EDAG: An End-to-End Document-level Framework for Chinese Financial Event Extraction

Doc2EDAG: An End-to-End Document-level Framework for Chinese Financial Event Extraction

Paper · 2019 · Shun Zheng, Wei Cao, Wei Xu et al.

Summary

Most event extraction (EE) systems only find event arguments within a single sentence. In financial documents, however, the arguments of one event are often spread over several sentences, and one document may describe several events. This work targets document-level EE (DEE) on Chinese financial announcements (ChFinAnn).

The authors propose Doc2EDAG, an end-to-end model. It recasts filling an event table as building an entity-based directed acyclic graph (EDAG) step by step, which breaks a hard table-filling problem into a sequence of easier path-expansion decisions. Doc2EDAG relies on two components:

  • a Transformer that encodes entities with document-wide context;
  • a memory mechanism that guides path expansion.

They also redefine the DEE task so that no trigger words are needed. This simplifies automatic labeling with distant supervision from an event knowledge base.

On a new large-scale ChFinAnn dataset, Doc2EDAG clearly beats the previous DCFEE pipeline and a greedy baseline, with the largest gains on documents containing several events. The method makes no domain-specific assumptions, so it could transfer to legal or medical texts.

Key points

  • Dataset: The new ChFinAnn dataset is about ten times larger than the one used for DCFEE, and roughly 30% of its documents contain multiple event records. Data and code are released at github.com/dolphin-zs/Doc2EDAG.
  • Event types and overall score: The dataset covers five event types (EF, ER, EU, EO, EP). Doc2EDAG reaches an average F1 of 76.3.
  • Gains over the strongest baseline, DCFEE-O:
    • By event type, Doc2EDAG improves F1 by 19.1 (EF), 4.2 (ER), 26.5 (EU), 28.4 (EO) and 13.4 (EP).
    • On the multi-event subset, the average gain is 17.7 F1.
  • Ablation results (average F1 drop when a component is removed):
    • path memory: −9.7;
    • scheduled sampling: −5.0;
    • document-level Transformer entity encoding: −2.1;
    • negative class weighting: −0.8.
  • Task formulation: The trigger-free design removes the need for predefined trigger dictionaries. Combining distant supervision with simple constraints already yields good document-level labels.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏢MicrosoftRedmond · United States240 works · 123,537 citations🎓Tsinghua UniversityBeijing · China137 works · 83,262 citations🏢Microsoft Research AsiaBeijing · China54 works · 59,921 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 1Research subjects and languages the work is about.Information extraction · 109 works

Cite

Cite as Shun Zheng, Wei Cao, Wei Xu and Jiang Bian. 2019. Doc2EDAG: An End-to-End Document-level Framework for Chinese Financial Event Extraction. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/d19-1032

@inproceedings{zheng2019doc2edag,
  title = {{Doc2EDAG: An End-to-End Document-level Framework for Chinese Financial Event Extraction}},
  author = {Shun Zheng, Wei Cao, Wei Xu and Jiang Bian},
  booktitle = {Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)},
  year = {2019},
  doi = {10.18653/v1/d19-1032},
  url = {https://doi.org/10.18653/v1/d19-1032}
}

References

22 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Mike Mintz, Steven Bills, Rion Snow, and Dan Jurafsky. 2009. Distant supervision for relation extraction without labeled data. In ACL. find
  2. David Ahn. 2006. The stages of event extraction. In Proceedings of the Workshop on Annotating and Reasoning about Time and Events.
  3. Thien Huu Nguyen, Kyunghyun Cho, and Ralph Grishman. 2016. Joint event extraction via recurrent neural networks. In NAACL.
  4. Samy Bengio, Oriol Vinyals, Navdeep Jaitly, and Noam Shazeer. 2015. Scheduled sampling for sequence prediction with recurrent neural networks. In NIPS. find
  5. Trung Minh Nguyen and Thien Huu Nguyen. 2019. One for all: Neural joint modeling of entities and events. In AAAI. find
  6. Yubo Chen, Shulin Liu, Xiang Zhang, Kang Liu, and Jun Zhao. 2017. Automatically labeled data generation for large scale event extraction. In ACL.
  7. Xiang Ren, Zeqiu Wu, Wenqi He, Meng Qu, Clare R Voss, Heng Ji, Tarek F Abdelzaher, and Jiawei Han. 2017. CoType: Joint extraction of typed entities and relations with knowledge bases. In WWW. find
  8. Yubo Chen, Liheng Xu, Kang Liu, Daojian Zeng, and Jun Zhao. 2015. Event extraction via dynamic multi-pooling convolutional neural networks. In ACL.
  9. Sebastian Riedel and Andrew McCallum. 2011. Fast and robust joint models for biomedical event extraction. In EMNLP. find
  10. Hang Yang, Yubo Chen, Kang Liu, Yang Xiao, and Jun Zhao. 2018. DCFEE: A document-level chinese financial event extraction system based on automatically labeled training data. In Proceedings of ACL 2018, System Demonstrations. find
  11. Sebastian Riedel, Limin Yao, and Andrew McCallum. 2010. Modeling relations and their mentions without labeled text. In ECML. find
  12. Lei Sha, Feng Qian, Baobao Chang, and Zhifang Sui. 2018. Jointly extracting event triggers and arguments by dependency-bridge rnn and tensor-based argument interaction. In AAAI. find
  13. Xiangrong Zeng, Daojian Zeng, Shizhu He, Kang Liu, and Jun Zhao. 2018a. Extracting relational facts by an end-to-end neural model with copy mechanism. In ACL.
  14. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In NIPS.
  15. Ying Zeng, Yansong Feng, Rong Ma, Zheng Wang, Rui Yan, Chongde Shi, and Dongyan Zhao. 2018b. Scale up event extraction learning via automatic training data generation. In AAAI. find
  16. Shaolei Wang, Yue Zhang, Wanxiang Che, and Ting Liu. 2018. Joint extraction of entities and relations based on a novel graph scheme. In IJCAI. find
  17. Tongtao Zhang and Heng Ji. 2018. Event extraction with generative adversarial imitation learning. arXiv preprint arXiv:1804.07881. — [arXiv:1804.07881](https://arxiv.org/abs/1804.07881) find
  18. Xiaozhi Wang, Xu Han, Zhiyuan Liu, Maosong Sun, and Peng Li. 2019. Adversarial training for weakly supervised event detection. In NAACL.
  19. Shun Zheng, Xu Han, Yankai Lin, Peilin Yu, Lu Chen, Ling Huang, Zhiyuan Liu, and Wei Xu. 2019. DIAG-NRE: A neural pattern diagnosis framework for distantly supervised neural relation extraction. In ACL. find
  20. Sen Wu, Luke Hsiao, Xiao Cheng, Braden Hancock, Theodoros Rekatsinas, Philip Levis, and Christopher Ré. 2018. Fonduer: Knowledge base construction from richly formatted data. In SIGMOD. find
  21. Suncong Zheng, Feng Wang, Hongyun Bao, Yuexing Hao, Peng Zhou, and Bo Xu. 2017. Joint extraction of entities and relations based on a novel tagging scheme. In ACL.
  22. Bishan Yang and Tom M. Mitchell. 2016. Joint extraction of events and entities within a document context. In NAACL.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zheng-2019-doc2edag-end-end-document-level/