More Data, More Relations, More Context and More Openness: A Review and Outlook for Relation Extraction

Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing, pages 745–758 December 4 - 7, 2020. c⃝2020 Association for Computational Linguistics 745 More Data, More Relations, More Context and More Openness: A Review and Outlook for Relation Extraction Xu Han1∗, Tianyu Gao2∗, Yankai Lin3∗, Hao Peng1, Yaoliang Yang1, Chaojun Xiao1, Zhiyuan Liu1† , Peng Li3, Maosong Sun1, Jie Zhou3 1Department of Computer Science and Technology, Tsinghua University, Beijing, China 2Princeton University, Princeton, NJ, USA 3Pattern Recognition Center, WeChat AI, Tencent Inc, China hanxu17@mails.tsinghua.edu.cn, tianyug@princeton.edu, yankailin@tencent.com Abstract Relational facts are an important component of human knowledge, which are hidden in vast amounts of text. In order to extract these facts from text, people have been working on rela- tion extraction (RE) for years. From early pat- tern matching to current neural networks, ex- isting RE methods have achieved significant progress. Yet with explosion of Web text and emergence of new relations, human knowl- edge is increasing drastically, and we thus re- quire “more” from RE: a more powerful RE system that can robustly utilize more data, ef- ficiently learn more relations, easily handle more complicated context, and flexibly gener- alize to more open domains. In this paper, we look back at existing RE methods, analyze key challenges we are facing nowadays, and show promising directions towards more powerful RE. We hope our view can advance this field and inspire more efforts in the community.1 1 Introduction Relational facts organize knowledge of the world in a triplet format. These structured facts act as an im- port role of human knowledge and are explicitly or implicitly hidden in the text. For example, “Steve Jobs co-founded Apple” indicates the fact (Apple Inc., founded by, Steve Jobs), and we can also infer the fact (USA, contains, New York) from “Hamilton made its debut in New York, USA”. As these structured facts could benefit down- stream applications, e.g, knowledge graph comple- tion (Bordes et al., 2013; Wang et al., 2014), search engine (Xiong et al., 2017; Schlichtkrull et al., 2018) and question answering (Bordes et al., 2014; Dong et al., 2015), many efforts have been devoted ∗ indicates equal contribution † Corresponding author e-mail: liuzy@tsinghua.edu.cn 1Most of the papers mentioned in this work are col- lected into the following paper list https://github. com/thunlp/NREPapers. to researching relation extraction (RE), which aims at extracting relational facts from plain text. More specifically, after identifying entity mentions (e.g., USA and New York) in text, the main goal of RE is to classify relations (e.g., contains) between these entity mentions from their context. The pioneering explorations of RE lie in statisti- cal approaches, such as pattern mining (Huffman, 1995; Califf and Mooney, 1997), feature-based methods (Kambhatla, 2004) and graphical models (Roth and Yih, 2002). Recently, with the develop- ment of deep learning, neural models have been widely adopted for RE (Zeng et al., 2014; Zhang et al., 2015) and achieved superior results. These RE methods …

Content truncated to 500 words