Joint Type Inference on Entities and Relations via Graph Convolutional Networks

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1361–1370 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 1361 Joint Type Inference on Entities and Relations via Graph Convolutional Networks Changzhi Sun 1, ∗, Yeyun Gong2, Yuanbin Wu1, 3, Ming Gong2, Daxing Jiang2, Man Lan1, Shiliang Sun1, and Nan Duan2 1Department of Computer Science and Technology, East China Normal University 2Microsoft Research Asia 3State Key Laboratory of Cognitive Intelligence, iFLYTEK {changzhisun}@stu.ecnu.edu.cn {ybwu,mlan,slsun}@cs.ecnu.edu.cn {yegong, nanduan, migon, djiang}@microsoft.com Abstract We develop a new paradigm for the task of joint entity relation extraction. It first identi- fies entity spans, then performs a joint infer- ence on entity types and relation types. To tackle the joint type inference task, we propose a novel graph convolutional network (GCN) running on an entity-relation bipartite graph. By introducing a binary relation classification task, we are able to utilize the structure of entity-relation bipartite graph in a more effi- cient and interpretable way. Experiments on ACE05 show that our model outperforms ex- isting joint models in entity performance and is competitive with the state-of-the-art in rela- tion performance. 1 Introduction Extracting entities and relations from plain texts is an important and challenging task in natural lan- guage processing. Given a sentence, the task aims to detect text spans with specific types (entities) and semantic relations among those text spans (re- lations). For example, in the Figure 1, “Toefting” is a person entity (PER), “teammates” is a person entity (PER), and the two entities have a Person- Social relation (PER-SOC). To tackle the task of entity relation extraction, various methods have been proposed, which can be divided into two categories: pipeline models and joint models. Pipeline models extract enti- ties and relations in two stages: entities are first extracted by an entity model, and then these ex- tracted entities are used as the inputs of a relation model. Pipeline models often ignore interactions between the two models and they suffer from error propagation. Joint models integrate information between entities and relations into a single model with the joint training, and have achieved better ∗ Work done while this author was an intern at Microsoft Research Asia. Toefting was convicted of assaulting a pair of wokers during a night out with national squad teammates in the capital … PER PER GPE Toefting teammates capital PER-SOC PHYS PHYS Toefting teammates capital (Toefting, teammates) (Toefting, capital) (teammates, capital) PER PER GPE PER-SOC-RIGHT PHYS-RIGHT PHYS-RIGHT Entity Types Entity Nodes Relation Nodes Relation Types Entity-Relation Graph Figure 1: An example from ACE05. The first part contains annotations and the second part is the entity- relation graph of the sentence used in GCN. results than the pipeline models. In this paper, we focus on joint models. More and more joint methods have been ap- plied to this task. Among them, Miwa and Bansal (2016); Katiyar and Cardie (2017) identify the en- tity with a sequence labelling model, and iden- tify the relation type with a multi-class …

Content truncated to 500 words