Denoising Relation Extraction from Document-level Distant Supervision

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 3683–3688, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 3683 Denoising Relation Extraction from Document-level Distant Supervision Chaojun Xiao1, Yuan Yao1, Ruobing Xie2, Xu Han1, Zhiyuan Liu1∗ Maosong Sun1, Fen Lin2, Leyu Lin2 1Department of Computer Science and Technology Institute for Artificial Intelligence, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology, China 2WeChat Search Application Department, Tencent, China {xcjthu,thu.hanxu13}@gmail.com, {yaoyuanthu,xrbsnowing}@163.com {lzy,sms}@tsinghua.edu.cn, {felicialin,goshawklin}@tencent.com Abstract Distant supervision (DS) has been widely used to generate auto-labeled data for sentence- level relation extraction (RE), which improves RE performance. However, the existing suc- cess of DS cannot be directly transferred to the more challenging document-level relation extraction (DocRE), since the inherent noise in DS may be even multiplied in document level and significantly harm the performance of RE. To address this challenge, we propose a novel pre-trained model for DocRE, which de- noises the document-level DS data via multi- ple pre-training tasks. Experimental results on the large-scale DocRE benchmark show that our model can capture useful information from noisy DS data and achieve promising results. The source code of this paper can be found in https://github.com/thunlp/DSDocRE. 1 Introduction Relation extraction (RE) aims to identify relational facts between entities from texts. Recently, neural relation extraction (NRE) models have been veri- fied in sentence-level RE (Zeng et al., 2014). Dis- tant supervision (DS) (Mintz et al., 2009) provides large-scale distantly-supervised data that multiplies instances and enables sufficient model training. Sentence-level RE focuses on extracting intra- sentence relations between entities in a sentence. However, it is extremely restricted with generality and coverage in practice, since there are plenty of inter-sentence relational facts hidden across multi- ple sentences. Statistics on a large-scale RE dataset constructed from Wikipedia documents show that at least 40.7% relational facts can only be inferred from multiple sentences (Yao et al., 2019). There- fore, document-level relation extraction (DocRE) is proposed to jointly extract both inter- and intra- sen- ∗Corresponding author. [1] Lebron James (born in Akron) is a profession- al basketball player. [2] Now James serves for the Lakers ... [8] James played basketball for St. Vincent–St. Mary High School in his hometown. Lakers Akron Lebron James member_of St. Vincent–St. Mary High School educated_at place_of_birth locate_in Figure 1: An example of DocRE. Given a document, DocRE models should capture the relational semantics across sentences to extract multiple relational facts. tence relations (Christopoulou et al., 2019). Fig. 1 gives a brief illustration of DocRE. Most DocRE models heavily rely on high-quality human-annotated training data, which is time- consuming and labor-intensive. However, it is ex- tremely challenging to extend the sentence-level DS to the document level. The challenges of con- ducting document-level DS mainly come from: (1) Each entity contains multiple mentions, and men- tions without relational context bring noise to entity representations; (2) The inherent noise of DS will be even multiplied at the document level. Statistics in Yao et al. (2019) show that 61.8% inter-sentence relation instances …

Content truncated to 500 words