Open Domain Event Extraction Using Neural Latent Variable Models

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2860–2871 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 2860 Open Domain Event Extraction Using Neural Latent Variable Models Xiao Liu1,2 and Heyan Huang1,2 and Yue Zhang3,4∗ 1School of Computer Science and Technology, Beijing Institute of Technology 2Zhejiang Lab, China {xiaoliu,hhy63}@bit.edu.cn 3School of Engineering, Westlake University 4Institute of Advanced Technology, Westlake Institute for Advanced Study yue.zhang@wias.org.cn Abstract We consider open domain event extraction, the task of extracting unconstraint types of events from news clusters. A novel latent variable neural model is constructed, which is scalable to very large corpus. A dataset is collected and manually annotated, with task-specific evalua- tion metrics being designed. Results show that the proposed unsupervised model gives better performance compared to the state-of-the-art method for event schema induction. 1 Introduction Extracting events from news text has received much research attention. The task typically con- sists of two subtasks, namely schema induction, which is to extract event templates that specify argument slots for given event types (Chambers, 2013; Cheung et al., 2013; Nguyen et al., 2015; Sha et al., 2016; Huang et al., 2016; Ahn, 2017; Yuan et al., 2018), and event extraction, which is to identify events with filled slots from a piece of news (Nguyen et al., 2016b; Sha et al., 2018; Liu et al., 2018a; Chen et al., 2018, 2015; Feng et al., 2016; Nguyen and Grishman, 2016; Liu et al., 2018b). Previous work focuses on extract- ing events from single news documents according to a set of pre-specified event types, such as arson, attack or earthquakes. While useful for tracking highly specific types of events from news, the above setting can be rel- atively less useful for decision making in secu- rity and financial markets, which can require com- prehensive knowledge on broad-coverage, fine- grained and dynamically-evolving event cate- gories. In addition, given the fact that different news agencies can report the same events, redun- dancy can be leveraged for better event extrac- tion. In this paper, we investigate open domain ∗Corresponding author. MUC 4 ODEE Document News Report News Report News Report News Cluster Trigger raise Agent UnitedHealth, UnitedHealth shares Patient 2018 forecast, better-than-expected profits, the insurance business Time the third quarter Variation 28% Unconstrained Types of Open Doamin Events with Their Own Schemas Trigger report Agent UnitedHealth Group, the largest U.S. health insurer Patient better-than-expected third-quarter earnings Time Tuesday Trigger predict Agent UnitedHealth Group Patient Medicare growth Type Perpetrator Instrument Target Victim Four Types of Events with Fixed Slots: Arson, Attack, Bombing and Kidnapping Figure 1: Comparison between MUC 4 and ODEE. event extraction (ODEE), which is to extract un- constraint types of events and induce universal event schemas from clusters of news reports. As shown in Figure 1, compared with tradi- tional event extraction task exemplified by MUC 4 (Sundheim, 1992), the task of ODEE poses ad- ditional challenges to modeling, which have not been considered in traditional methods. First, more than …

Content truncated to 500 words