Attention Guided Graph Convolutional Networks for Relation Extraction

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 241–251 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 241 Attention Guided Graph Convolutional Networks for Relation Extraction Zhijiang Guo∗, Yan Zhang∗ and Wei Lu StatNLP Research Group Singapore University of Technology and Design {zhijiang guo,yan zhang}@mymail.sutd.edu.sg, luwei@sutd.edu.sg Abstract Dependency trees convey rich structural in- formation that is proven useful for extract- ing relations among entities in text. However, how to effectively make use of relevant infor- mation while ignoring irrelevant information from the dependency trees remains a challeng- ing research question. Existing approaches employing rule based hard-pruning strategies for selecting relevant partial dependency struc- tures may not always yield optimal results. In this work, we propose Attention Guided Graph Convolutional Networks (AGGCNs), a novel model which directly takes full dependency trees as inputs. Our model can be understood as a soft-pruning approach that automatically learns how to selectively attend to the relevant sub-structures useful for the relation extrac- tion task. Extensive results on various tasks including cross-sentence n-ary relation extrac- tion and large-scale sentence-level relation ex- traction show that our model is able to bet- ter leverage the structural information of the full dependency trees, giving significantly bet- ter results than previous approaches. 1 Introduction Relation extraction aims to detect relations among entities in the text. It plays a significant role in a variety of natural language processing applica- tions including biomedical knowledge discovery (Quirk and Poon, 2017), knowledge base popula- tion (Zhang et al., 2017) and question answering (Yu et al., 2017). Figure 1 shows an example about expressing a relation sensitivity among three enti- ties L858E, EGFR and gefitinib in two sentences. Most existing relation extraction models can be categorized into two classes: sequence-based and dependency-based. Sequence-based models op- erate only on the word sequences (Zeng et al., ∗Equally Contributed. 2014; Wang et al., 2016), whereas dependency- based models incorporate dependency trees into the models (Bunescu and Mooney, 2005; Peng et al., 2017). Compared to sequence-based mod- els, dependency-based models are able to capture non-local syntactic relations that are obscure from the surface form alone (Zhang et al., 2018). Var- ious pruning strategies are also proposed to distill the dependency information in order to further im- prove the performance. Xu et al. (2015b,c) apply neural networks only on the shortest dependency path between the entities in the full tree. Miwa and Bansal (2016) reduce the full tree to the subtree below the lowest common ancestor (LCA) of the entities. Zhang et al. (2018) apply graph convolu- tional networks (GCNs) (Kipf and Welling, 2017) model over a pruned tree. This tree includes to- kens that are up to distance K away from the de- pendency path in the LCA subtree. However, rule-based pruning strategies might eliminate some important information in the full tree. Figure 1 shows an example in cross-sentence n-ary relation extraction that the key tokens par- tial response would be excluded if the model only takes the pruned …

Content truncated to 500 words