Double Graph Based Reasoning for Document-level Relation Extraction

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 1630–1640, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 1630 Double Graph Based Reasoning for Document-level Relation Extraction Shuang Zeng1,2∗, Runxin Xu1∗, Baobao Chang1,2† and Lei Li3 1Key Laboratory of Computational Linguistics, Peking University, MOE, China 2School of Software and Microelectronics, Peking University, China 3ByteDance AI Lab {zengs,chbb}@pku.edu.cn runxinxu@gmail.com lileilab@bytedance.com Abstract Document-level relation extraction aims to ex- tract relations among entities within a docu- ment. Different from sentence-level relation extraction, it requires reasoning over multiple sentences across paragraphs. In this paper, we propose Graph Aggregation-and-Inference Network (GAIN), a method to recognize such relations for long paragraphs. GAIN con- structs two graphs, a heterogeneous mention- level graph (MG) and an entity-level graph (EG). The former captures complex interac- tion among different mentions and the latter aggregates mentions underlying for the same entities. Based on the graphs we propose a novel path reasoning mechanism to infer re- lations between entities. Experiments on the public dataset, DocRED, show GAIN achieves a significant performance improvement (2.85 on F1) over the previous state-of-the-art. Our code is available at https://github.com/ PKUnlp-icler/GAIN. 1 Introduction The task of identifying semantic relations between entities from text, namely relation extraction (RE), plays a crucial role in a variety of knowledge-based applications, such as question answering (Yu et al., 2017, Qiu et al., 2019) and large-scale knowledge graph construction. Previous methods (Zeng et al., 2014; Zeng et al., 2015; Xiao and Liu, 2016; Zhang et al., 2017; Zhang et al., 2018; Baldini Soares et al., 2019) focus on sentence-level RE, which pre- dicts relations among entities in a single sentence. However, sentence-level RE models suffer from an inevitable limitation – they fail to recognize re- lations between entities across sentences. Hence, extracting relations at the document-level is neces- sary for a holistic understanding of knowledge in text. ∗Equal contribution. †Corresponding author. Elias Brown [1] Elias Brown (May 9, 1793– July 7, 1857) was a U.S. Representative from Maryland. [2] Born near Baltimore, Maryland, Brown attended the common schools. … [7] He died near Baltimore, Maryland, and is interred in a private cemetery near Eldersburg, Maryland. Subject: Maryland Object: U.S. relation: country Subject: Baltimore；Eldersburg Object: Maryland relation: located in the administrative territorial entity Subject: Baltimore；Eldersburg Object: U.S. relation: country Figure 1: An example document and its desired rela- tions from DocRED (Yao et al., 2019). Entity men- tions and relations involved in these relation instances are colored. Other mentions are underlined for clarity. There are several major challenges in effective relation extraction at the document-level. Firstly, the subject and object entities involved in a relation may appear in different sentences. Therefore a re- lation cannot be identified based solely on a single sentence. Secondly, the same entity may be men- tioned multiple times in different sentences. Cross- sentence context information has to be aggregated to represent the entity better. Thirdly, the identifi- cation of many relations requires techniques of log- ical reasoning. This means these relations can only be …

Content truncated to 500 words