Adaptive Knowledge-Enhanced Bayesian Meta-Learning for Few-shot Event Detection

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 2417–2429 August 1–6, 2021. ©2021 Association for Computational Linguistics 2417 Adaptive Knowledge-Enhanced Bayesian Meta-Learning for Few-shot Event Detection Shirong Shen1 and Tongtong Wu1 and Guilin Qi1∗ and Yuan-Fang Li2 and Gholamreza Haffari 2 and Sheng Bi1 1School of Computer Science and Engineering, Southeast University, China 2Faculty of Information Technology, Monash University, Melbourne, Australia {ssr, wutong8023, gqi}@seu.edu.cn, yuanfang.li@monash.edu, Gholamreza.Haffari@monash.edu, bisheng@seu.edu.cn Abstract Event detection (ED) aims at detecting event trigger words in sentences and classifying them into specific event types. In real-world applications, ED typically does not have suf- ficient labelled data, thus can be formulated as a few-shot learning problem. To tackle the issue of low sample diversity in few-shot ED, we propose a novel knowledge-based few- shot event detection method which uses a definition-based encoder to introduce external event knowledge as the knowledge prior of event types. Furthermore, as external knowl- edge typically provides limited and imper- fect coverage of event types, we introduce an adaptive knowledge-enhanced Bayesian meta- learning method to dynamically adjust the knowledge prior of event types. Experiments show our method consistently and substan- tially outperforms a number of baselines by at least 15 absolute F1 points under the same few- shot settings. 1 Introduction Event detection is an important task in informa- tion extraction, aiming at detecting event triggers from text and then classifying them into event types (Chen et al., 2015). For example, in “The po- lice arrested Harry on charges of manslaughter”, the trigger word is arrested, indicating an “Arrest” event. Event detection has been widely applied in Twitter analysis (Zhou et al., 2017), legal case ex- traction (de Araujo et al., 2017), and financial event extraction (Zheng et al., 2019), to name a few. Typical approaches to event detection (Chen et al., 2015; McClosky et al., 2011; Liu et al., 2019) generally rely on large-scale annotated datasets for training. Yet in real-world applications, adequate labeled data is usually unavailable. Hence, meth- ods that generalize effectively with small quantities * Corresponding author. Figure 1: A 3-way 3-shot event detection example, in which the model uses the support set to predict the event types of query samples. of labeled samples and adapt quickly to new event types are highly desirable for event detection. Various approaches have been proposed to en- able learning from only a few samples, i.e., few- shot learning (Finn et al., 2017; Snell et al., 2017; Zhang et al., 2018a). Yet few-shot event detection (FSED) has been less studied until recently (Lai et al., 2020a; Deng et al., 2020). Although these methods achieve encouraging progress on typical N-way M-shot setting (Figure 1), the performance remains unsatisfactory as the diversity of examples in the support set is usually limited. Intuitively, introducing high-quality semantic knowledge, such as FrameNet (Baker et al., 1998), is a potential solution to the insufficient diversity issue (Qu et al., 2020; Tong et al., 2020; Liu et al., 2016, 2020). However, as shown in Figure 2, such knowledge-enhanced methods also …

Content truncated to 500 words