AI helps you reading Science
AI generates interpretation videos
AI extracts and analyses the key points of the paper to generate videos automatically
AI parses the academic lineage of this thesis
AI extracts a summary of this paper
We demonstrate how gating mechanisms, gate diversity, and graph structure can be used to integrating syntactic information and improve the hidden vectors for event detection models
Event Detection: Gate Diversity and Syntactic Importance Scores for Graph Convolution Neural Networks
EMNLP 2020, pp.5405-5411, (2020)
Recent studies on event detection (ED) have shown that the syntactic dependency graph can be employed in graph convolution neural networks (GCN) to achieve state-of-the-art performance. However, the computation of the hidden vectors in such graph-based models is agnostic to the trigger candidate words, potentially leaving irrelevant infor...More
PPT (Upload PPT)
- Event Detection (ED) is an important task in Information Extraction of Natural Language Processing.
- The main goal of this task is to identify event instances presented in text.
- The event detection task, precisely speaking, seeks to identify the event triggers and classify them into some types of interest.
- (1) They’ll be fired on at the crossing.
- (2) She is on her way to get fired.
- An ideal ED system should be able to recognize the two words “fired” in the sentences as the triggers of the event types “Attack” and “End-Position”
- Event Detection (ED) is an important task in Information Extraction of Natural Language Processing
- We propose to filter the noisy information from the hidden vectors of graph convolution neural net-works (GCN) so that only the relevant information for the trigger candidate is preserved
- Datasets: We evaluate our proposed model on two event detection (ED) datasets, i.e., ACE2005 and Litbank
- The first class includes the models with noncontextualized embedding, i.e., CNN: a CNN model (Nguyen and Grishman, 2015), NCNN: non-consecutive CNN model: (Nguyen and Grishman, 2016), and GCN-ED: a GCN model (Nguyen and Grishman, 2018)
- We demonstrate how gating mechanisms, gate diversity, and graph structure can be used to integrating syntactic information and improve the hidden vectors for ED models
- The proposed model achieves state-of-the-art performance on two ED datasets
- Datasets: The authors evaluate the proposed model on two ED datasets, i.e., ACE2005 and Litbank.
- Once the authors re-implement this model and apply it to the data version pre-processed and provided by the prior work (Nguyen and Grishman, 2015, 2018), the authors are only able to achieve an F1 score of 76.2% on the test set.
- In this work, the authors employ the exact data version that has been pre-processed and released by the early work on ED for ACE-2005 in (Nguyen and Grishman, 2015, 2018) and for Litbank in (Sims et al, 2019)
- The authors compare the model with two classes of baselines on ACE-2005. The first class includes the models with noncontextualized embedding, i.e., CNN: a CNN model (Nguyen and Grishman, 2015), NCNN: non-consecutive CNN model: (Nguyen and Grishman, 2016), and GCN-ED: a GCN model (Nguyen and Grishman, 2018).
- The second class of baselines concern the models with the contextualized embeddings, i.e., DMBERT: a model with dynamic pooling (Wang et al, 2019) and BERT+MLP: a MLP model with BERT (Yang et al, 2019)
- These models currently have the best-reported performance for ED on ACE-2005.
- For Litbank, the authors use the following baselines reported in the original paper (Sims et al, 2019): BiLSTM: a BiLSTM model with word2vec, BERT+BiLSTM: a BiLSTM model with BERT, and DMBERT (Wang et al, 2019)
- The authors demonstrate how gating mechanisms, gate diversity, and graph structure can be used to integrating syntactic information and improve the hidden vectors for ED models.
- The proposed model achieves state-of-the-art performance on two ED datasets.
- The authors plan to apply the proposed model for the related tasks and other settings of ED, including new type extension (Nguyen et al, 2016b; Lai and Nguyen, 2019), and few-shot learning (Lai et al, 2020a,b)
- Table1: Performance on the ACE-2005 test set
- Table2: Performance on the Litbank test set
- Table3: Ablation study on the ACE-2005 dev set
- Prior studies on ED involve handcrafted feature engineering for statistical models (Ahn, 2006; Ji and Grishman, 2008; Hong et al, 2011; Li et al, 2013; Mitamura et al, 2015) and deep neural networks, e.g., CNN (Chen et al, 2015, 2017; Nguyen and Grishman, 2015; Nguyen et al, 2016g), RNN (Nguyen et al, 2016; Jagannatha and Yu, 2016; Feng et al, 2016), attention mechanism (Liu et al, 2017; Chen et al, 2018), contextualized embeddings (Yang et al, 2019), and adversarial training (Wang et al, 2019). The last few years witness the success of graph convolutional neural networks for ED (Nguyen and Grishman, 2018; Liu et al, 2018; Veyseh et al, 2019; Yan et al, 2019) where the dependency trees are employed to boost the performance. However, these graph-based models have not considered representation regulation for GCNs and exploiting graph-based distances as we do in this work.
Task Description: The goal of ED consists of identifying trigger words (trigger identification) and classifying them for the event types of interest (event classification). Following the previous studies (Nguyen and Grishman, 2015), we combine these two tasks as a single multi-way classification task by introducing a None class, indicating non-event. Formally, given a sentence X = [x1, x2, . . . , xn] of n words, and an index t (1 ≤ t ≤ n) of the trigger candidate xt, the goal is to predict the event type y∗ for the candidate xt. Our ED model consists of three modules: (1) Sentence Encoder, (2) GCN and Gate Diversity, and (3) Graph and Model Consistency.
- This research has been supported in part by Vingroup Innovation Foundation (VINIF) in project code VINIF.2019.DA18 and Adobe Research Gift
- This research is also based upon work supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via IARPA Contract No 2019-19051600006 under the Better Extraction from Text Towards Enhanced Retrieval (BETTER) Program
- David Ahn. 2006. The stages of event extraction. In Proceedings of the Workshop on Annotating and Reasoning about Time and Events.
- Yubo Chen, Shulin Liu, Xiang Zhang, Kang Liu, and Jun Zhao. 2017. Automatically labeled data generation for large scale event extraction. In ACL.
- Yubo Chen, Liheng Xu, Kang Liu, Daojian Zeng, and Jun Zhao. 2015. Event extraction via dynamic multipooling convolutional neural networks. In ACLIJCNLP.
- Yubo Chen, Hang Yang, Kang Liu, Jun Zhao, and Yantao Jia. 2018. Collective event detection via a hierarchical and bias tagging networks with gated multilevel attention mechanisms. In EMNLP.
- Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In NAACL:HLT, pages 4171–4186.
- Xiaocheng Feng, Lifu Huang, Duyu Tang, Heng Ji, Bing Qin, and Ting Liu. 201A languageindependent neural network for event detection. In ACL (Volume 2: Short Papers), volume 2, pages 66– 71.
- Yu Hong, Jianfeng Zhang, Bin Ma, Jianmin Yao, Guodong Zhou, and Qiaoming Zhu. 2011. Using cross-entity inference to improve event extraction. In ACL.
- Abhyuday N Jagannatha and Hong Yu. 2016. Bidirectional rnn for medical event detection in electronic health records. In NAACL.
- Heng Ji and Ralph Grishman. 2008. Refining event extraction through cross-document inference. In ACL.
- Thomas N. Kipf and Max Welling. 2017. Semisupervised classification with graph convolutional networks. In ICLR.
- Viet Dac Lai, Franck Dernoncourt, and Thien Huu Nguyen. 2020a. Exploiting the matching information in the support set for few shot event classification. In Proceedings of the 24th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD).
- Viet Dac Lai, Franck Dernoncourt, and Thien Huu Nguyen. 2020b. Extensively matching for few-shot learning event detection. In Proceedings of the 1st Joint Workshop on Narrative Understanding, Storylines, and Events (NUSE) at ACL 2020.
- Viet Dac Lai and Thien Nguyen. 2019. Extending event detection to new types with learning from keywords. In Proceedings of the 5th Workshop on Noisy User-generated Text (W-NUT 2019) at EMNLP 2019.
- Qi Li, Heng Ji, and Liang Huang. 2013. Joint event extraction via structured prediction with global features. In ACL.
- Shulin Liu, Yubo Chen, Kang Liu, and Jun Zhao. 2017. Exploiting argument information to improve event detection via supervised attention mechanisms. In ACL.
- Xiao Liu, Zhunchen Luo, and He-Yan Huang. 2018. Jointly multiple events extraction via attentionbased graph information aggregation. In EMNLP, pages 1247–1256.
- Diego Marcheggiani and Ivan Titov. 20Encoding sentences with graph convolutional networks for semantic role labeling. In EMNLP.
- Teruko Mitamura, Zhengzhong Liu, and Eduard Hovy. 2015. Overview of tac kbp 2015 event nugget track. In TAC.
- Thien Huu Nguyen, Kyunghyun Cho, and Ralph Grishman. 2016. Joint event extraction via recurrent neural networks. In NAACL.
- Thien Huu Nguyen, Lisheng Fu, Kyunghyun Cho, and Ralph Grishman. 2016b. A two-stage approach for extending event detection to new types via neural networks. In Proceedings of the 1st ACL Workshop on Representation Learning for NLP (RepL4NLP).
- Thien Huu Nguyen and Ralph Grishman. 2015. Event detection and domain adaptation with convolutional neural networks. In ACL-IJCNLP.
- Thien Huu Nguyen and Ralph Grishman. 2016. Modeling skip-grams for event detection with convolutional neural networks. In EMNLP.
- Thien Huu Nguyen and Ralph Grishman. 2018. Graph convolutional networks with argument-aware pooling for event detection. In AAAI.
- Thien Huu Nguyen, Adam Meyers, and Ralph Grishman. 2016g. New york university 2016 system for kbp event nugget: A deep learning approach. In Proceedings of Text Analysis Conference (TAC).
- Trung Minh Nguyen and Thien Huu Nguyen. 2019. One for all: Neural joint modeling of entities and events. In AAAI.
- Matthew Sims, Jong Ho Park, and David Bamman. 2019. Literary event detection. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3623–3634.
- Amir Pouran Ben Veyseh, Thien Huu Nguyen, and Dejing Dou. 2019. Graph based neural networks for event factuality prediction using syntactic and semantic structures. In ACL.
- Christopher Walker, Stephanie Strassel, Julie Medero, and Kazuaki Maeda. 2006. Ace 2005 multilingual training corpus. Linguistic Data Consortium, Philadelphia, 57.
- Xiaozhi Wang, Xu Han, Zhiyuan Liu, Maosong Sun, and Peng Li. 2019. Adversarial training for weakly supervised event detection. In NAACL-HLT, pages 998–1008.
- Haoran Yan, Xiaolong Jin, Xiangbin Meng, Jiafeng Guo, and Xueqi Cheng. 2019. Event detection with multi-order graph convolution and aggregated attention. In EMNLP-IJCNLP, pages 5770–5774.
- Sen Yang, Dawei Feng, Linbo Qiao, Zhigang Kan, and Dongsheng Li. 2019. Exploring pre-trained language models for event extraction and generation. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5284– 5294.