Document-level Event Extraction via Heterogeneous Graph-based Interaction Model with a Tracker

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 3533–3546 August 1–6, 2021. ©2021 Association for Computational Linguistics 3533 Document-level Event Extraction via Heterogeneous Graph-based Interaction Model with a Tracker Runxin Xu1, Tianyu Liu1, Lei Li3 and Baobao Chang1,2∗ 1Key Laboratory of Computational Linguistics, Peking University, MOE, China 2Peng Cheng Laboratory, Shenzhen, China 3ByteDance AI Lab runxinxu@gmail.com,lileilab@bytedance.com {tianyu0421,chbb}@pku.edu.cn Abstract Document-level event extraction aims to rec- ognize event information from a whole piece of article. Existing methods are not effective due to two challenges of this task: a) the tar- get event arguments are scattered across sen- tences; b) the correlation among events in a document is non-trivial to model. In this pa- per, we propose Heterogeneous Graph-based Interaction Model with a Tracker (GIT) to solve the aforementioned two challenges. For the first challenge, GIT constructs a hetero- geneous graph interaction network to capture global interactions among different sentences and entity mentions. For the second, GIT in- troduces a Tracker module to track the ex- tracted events and hence capture the interde- pendency among the events. Experiments on a large-scale dataset (Zheng et al., 2019) show GIT outperforms the existing best methods by 2.8 F1. Further analysis reveals GIT is ef- fective in extracting multiple correlated events and event arguments that scatter across the document. Our code is available at https: //github.com/RunxinXu/GIT. 1 Introduction Event Extraction (EE) is one of the key and chal- lenging tasks in Information Extraction (IE), which aims to detect events and extract their arguments from the text. Most previous methods (Chen et al., 2015; Nguyen et al., 2016; Liu et al., 2018; Yang et al., 2019; Du and Cardie, 2020b) focus on sentence-level EE, extracting events from a sin- gle sentence. The sentence-level model, however, fails to extract events whose arguments spread in multiple sentences, which is much more common in real-world scenarios. Hence, extracting events at the document-level is critical. It has attracted much attention recently (Yang et al., 2018; Zheng et al., 2019; Du and Cardie, 2020a; Du et al., 2020). *Corresponding author. [1] On Nov 6, 2014, the company received a letter of share reduction from Mingting Wu, the shareholder of the company. [2] Mingting Wu decreased his holding of 7.2 million shares of the company on the Shenzhen Stock Exchange on Nov 6, 2014. [3] The 7.2 million shares of the company Mingting Wu reduced this time were transferred to Xiaoting Wu. [4] Xiaoting Wu is the daughter of Mingting Wu, and they were identified as persons acting in concert according to relevant regulations. EventTypeEquityHolderTradedShares StartDate 7.2 million Nov 6, 2014 Xiaoting Wu EO EU Mingting Wu 7.2 million Nov 6, 2014 … … … Figure 1: An example document from a Chinese dataset proposed by Zheng et al. (2019) in the financial domain, and we translate it into English for illustration. Entity mentions are colored. Due to space limitation, we only show four associated sentences and three …

Content truncated to 500 words