OntoED: Low-resource Event Detection with Ontology Embedding

OntoED: Low-resource Event Detection with Ontology Embedding Shumin Deng1,2∗, Ningyu Zhang1,2∗, Luoqiu Li1,2, Hui Chen3, Huaixiao Tou3, Mosha Chen3, Fei Huang3, Huajun Chen1,2† 1Zhejiang University & AZFT Joint Lab for Knowledge Engine, China 2Hangzhou Innovation Center, Zhejiang University, China 3Alibaba Group, China {231sm, zhangningyu, 3160102409}@zju.edu.cn, {weidu.ch, huaixiao.thx}@alibaba-inc.com, {chenmosha.cms, f.huang}@alibaba-inc.com, huajunsir@zju.edu.cn Abstract Event Detection (ED) aims to identify event trigger words from a given text and classify it into an event type. Most of current methods to ED rely heavily on training instances, and almost ignore the correlation of event types. Hence, they tend to suffer from data scarcity and fail to handle new unseen event types. To address these problems, we formulate ED as a process of event ontology population: linking event instances to pre-defined event types in event ontology, and propose a novel ED frame- work entitled OntoED with ontology embed- ding. We enrich event ontology with linkages among event types, and further induce more event-event correlations. Based on the event ontology, OntoED can leverage and propagate correlation knowledge, particularly from data- rich to data-poor event types. Furthermore, OntoED can be applied to new unseen event types, by establishing linkages to existing ones. Experiments indicate that OntoED is more pre- dominant and robust than previous approaches to ED, especially in data-scarce scenarios. 1 Introduction Event Detection (ED) (Chen et al., 2015) is the task to extract structure information of events from un- structured texts. For example, in the event mention “Jack is married to the Iraqi microbiologist known as Dr. Germ.”, an ED model should identify the event type as ‘Marry’ where the word ‘married’ triggers the event. The extracted events with canon- ical structure facilitate various social applications, such as biomedical science (Li et al., 2019; Wang et al., 2020c), financial analysis (Deng et al., 2019; Liang et al., 2020), fake news detection (Wang et al., 2018; Nikiforos et al., 2020) and so on. As a non-trivial task, ED suffers from the low- resource issues. On the one hand, the maldistribu- ∗ Equal Contribution. † Corresponding Author. Sentence (321) Acquit (27) Pardon (10) Justice Attack (4816) Riot (30) Conflict Injure (547) Marry (26229) Life Divorce (65) Cause Cause Before Before Before SubSuper CoSuper CoSuper New Event Type Demonstrate Be-Born New Event Type Figure 1: Low-resource Event Detection w.r.t. Event Correlation in FewEvent (Deng et al., 2020) Dataset. tion of samples is quite serious in ED benchmark datasets, e.g., FewEvent (Deng et al., 2020) and MAVEN (Wang et al., 2020b), where a large por- tion of event types contain relatively few training instances. As shown in Figure 1, the sample size of two event types Attack and Riot differs greatly (4816 & 30). In low-resource scenarios, supervised ED models (Chen et al., 2015; Nguyen et al., 2016; Liu et al., 2018) are prone to overfitting since they require sufficient training instances for all event types. On the other hand, real-world applications tend to be open and evolve promptly, and accord- ingly there can be numerous new unseen event types. …

Content truncated to 500 words