Entity-centered Cross-document Relation Extraction

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 9871 - 9881 December 7-11, 2022 ©2022 Association for Computational Linguistics Entity-centered Cross-document Relation Extraction Fengqi Wang1, Fei Li1∗, Hao Fei2, Jingye Li1, Shengqiong Wu2, Fangfang Su1, Wenxuan Shi1, Donghong Ji1 and Bo Cai1* 1 Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, China 2 School of Computing, National University of Singapore, Singapore {wangfengqi,lifei_csnlp,theodorelee,fangsu,shiwenxuan,dhji,caib}@whu.edu.cn haofei37@nus.edu.sg, swu@u.nus.edu Abstract Relation Extraction (RE) is a fundamental task of information extraction, which has attracted a large amount of research attention. Previous studies focus on extracting the relations within a sentence or document, while currently re- searchers begin to explore cross-document RE. However, current cross-document RE methods directly utilize text snippets surrounding the tar- get entities in multiple given documents, which brings considerable noisy and non-relevant sen- tences. Moreover, they utilize all the text paths in a document bag in a coarse-grained way, without considering the connections between these text paths. In this paper, we aim to ad- dress both of these shortages and push the state- of-the-art for cross-document RE. First, we fo- cus on input construction for our RE model and propose an entity-based document-context filter to retain useful information in the given docu- ments by using the bridge entities in the text paths. Second, we propose a cross-document RE model based on cross-path entity relation at- tention, which allows the entity relations across text paths to interact with each other. We com- pare our cross-document RE method with the state-of-the-art methods in the dataset CodRED. Our method outperforms them by at least 10% in F1, thus demonstrating its effectiveness.1 1 Introduction Relation Extraction (RE) aims to detect the seman- tic relations between a pair of target entities in a given text, which has long been a fundamental task in natural language processing (NLP). Most of RE studies are under the assumption that en- tity pairs are within a sentence (i.e., sentence-level RE) (Zeng et al., 2014; dos Santos et al., 2015; Cai et al., 2016; Zhou et al., 2016; Zhang et al., 2018; Fei et al., 2021e) or a document (i.e., document- level RE) (Christopoulou et al., 2019; Nan et al., ∗ Corresponding author 1Code: https://github.com/MakiseKuurisu/ecrim 2020; Zeng et al., 2020; Li et al., 2021; Fei et al., 2022a; Zhang et al., 2021b). Another line considers the research of cross-text RE, where entity pairs are separated into different text units, (i.e., cross- sentence RE or N-ary RE) (Peng et al., 2017). The latest RE research has moved to cross- document RE (CodRE), i.e., the target entities are located in different documents (Yao et al., 2021). As exemplified in Fig. 1, a CodRE model needs to first retrieve the relevant documents and then recognizes the key text paths in these documents for relation reasoning. In Yao et al. (2021), the task is formalized based on the idea of distant supervi- sion (Mintz et al., 2009), i.e., the text …

Content truncated to 500 words