Graph Neural Networks with Generated Parameters for Relation Extraction

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1331–1339 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 1331 Graph Neural Networks with Generated Parameters for Relation Extraction Hao Zhu1 Yankai Lin1 Zhiyuan Liu1 Jie Fu2 Tat-seng Chua3 Maosong Sun1 1 State Key Lab on Intelligent Technology and Systems Department of Computer Science and Technology Institute for Artificial Intelligence, Tsinghua Univerisity, Beijing, China 2 National University of Singapore, Singapore 3 Universit´e de Montr´eal, Montr´eal, Qu´ebec, Canada {{zhuhao15,linyk14}@mails,{liuzy, sms}@}tsinghua.edu.cn jie.fu@polymtl.ca,chuats@comp.nus.edu.sg Abstract In this paper, we propose a novel graph neu- ral network with generated parameters (GP- GNNs). The parameters in the propagation module, i.e. the transition matrices used in message passing procedure, are produced by a generator taking natural language sentences as inputs. We verify GP-GNNs in relation ex- traction from text, both on bag- and instance- settings. Experimental results on a human- annotated dataset and two distantly supervised datasets show that multi-hop reasoning mecha- nism yields significant improvements. We also perform a qualitative analysis to demonstrate that our model could discover more accu- rate relations by multi-hop relational reason- ing. Codes and data are released at https: //github.com/thunlp/gp-gnn. 1 Introduction In recent years, graph neural networks (GNNs) have been applied to various fields of ma- chine learning, including node classification (Kipf and Welling, 2016), relation classification (Schlichtkrull et al., 2017), molecular property prediction (Gilmer et al., 2017), few-shot learning (Garcia and Bruna, 2018), and achieved promising results on these tasks. These works have demon- strated GNNs’ strong power to process relational reasoning on graphs. Relational reasoning aims to abstractly reason about entities/objects and their relations, which is an important part of human intelligence. Be- sides graphs, relational reasoning is also of great importance in many natural language processing The original idea comes from several discussions be- tween Hao Zhu and Jie Fu while Hao Zhu visiting NUS; Hao Zhu designed research, prepared datasets, and conducted experiments; Jie Fu, Yankai Lin, and Zhiyuan Liu also par- ticipated in discussion while planning experiments; Zhiyuan Liu, Tat-seng Chua and Maosong Sun proofread the paper. Zhiyuan Liu serves as the corresponding author. tasks such as question answering, relation extrac- tion, summarization, etc. Consider the example shown in Fig. 1, existing relation extraction mod- els could easily extract the facts that Luc Besson directed a film L´eon: The Professional and that the film is in English, but fail to infer the rela- tionship between Luc Besson and English with- out multi-hop relational reasoning. By consider- ing the reasoning patterns, one can discover that Luc Besson could speak English following a rea- soning logic that Luc Besson directed L´eon: The Professional and this film is in English indicates Luc Besson could speak English. However, most existing GNNs can only process multi-hop rela- tional reasoning on pre-defined graphs and cannot be directly applied in natural language relational reasoning. Enabling multi-hop relational reason- ing in natural languages remains an open problem. To address this issue, in this paper, we …

Content truncated to 500 words