Global-to-Local Neural Networks for Document-Level Relation Extraction

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 3711–3721, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 3711 Global-to-Local Neural Networks for Document-Level Relation Extraction Difeng Wang† Wei Hu†, ‡, ∗ Ermei Cao† Weijian Sun§ † State Key Laboratory for Novel Software Technology, Nanjing University, China ‡ National Institute of Healthcare Data Science, Nanjing University, China § Huawei Technologies Co., Ltd. {dfwang,emcao}.nju@gmail.com, whu@nju.edu.cn, sunweijian@huawei.com Abstract Relation extraction (RE) aims to identify the semantic relations between named entities in text. Recent years have witnessed it raised to the document level, which requires complex reasoning with entities and mentions through- out an entire document. In this paper, we pro- pose a novel model to document-level RE, by encoding the document information in terms of entity global and local representations as well as context relation representations. Entity global representations model the semantic in- formation of all entities in the document, en- tity local representations aggregate the contex- tual information of multiple mentions of spe- cific entities, and context relation representa- tions encode the topic information of other re- lations. Experimental results demonstrate that our model achieves superior performance on two public datasets for document-level RE. It is particularly effective in extracting relations between entities of long distance and having multiple mentions. 1 Introduction Relation extraction (RE) aims to identify the se- mantic relations between named entities in text. While previous work (Zeng et al., 2014; Zhang et al., 2015, 2018) focuses on extracting relations within a sentence, a.k.a. sentence-level RE, recent studies (Verga et al., 2018; Christopoulou et al., 2019; Sahu et al., 2019; Yao et al., 2019) have esca- lated it to the document level, since a large amount of relations between entities usually span across multiple sentences in the real world. According to an analysis on Wikipedia corpus (Yao et al., 2019), at least 40.7% of relations can only be extracted on the document level. Compared with sentence-level RE, document- level RE requires more complex reasoning, such ∗Corresponding author [S1] Pacific Fair is a major shopping centre in Broadbeach Waters on the Gold Coast, Queensland, Australia. [S11] Pacific Fair fronts Little Tallebudgera Creek and is the southern end of the Surfers Riverwalk. … … ? coreference Figure 1: An example of document-level RE excerpted from the DocRED dataset (Yao et al., 2019). Arrows denote intra/inter-sentential relations. as logical reasoning, coreference reasoning and common-sense reasoning. A document often con- tains many entities, and some entities have multi- ple mentions under the same phrase of alias. To identify the relations between entities appearing in different sentences, document-level RE models must be capable of modeling the complex interac- tions between multiple entities and synthesizing the context information of multiple mentions. Figure 1 shows an example of document-level RE. Assume that one wants to extract the relation between “Surfers Riverwalk” in S11 and “Queens- land” in S1. One has to find that “Surfers River- walk” contains “Pacific Fair” (from S11), and “Pa- cific Fair” (coreference) is located in “Queensland” (from …

Content truncated to 500 words