Open Relation Extraction: Relational Knowledge Transfer from Supervised Data to Unsupervised Data

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 219–228, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 219 Open Relation Extraction: Relational Knowledge Transfer from Supervised Data to Unsupervised Data Ruidong Wu1∗, Yuan Yao1∗, Xu Han1, Ruobing Xie2, Zhiyuan Liu1† , Fen Lin2, Leyu Lin2, Maosong Sun1 1Department of Computer Science and Technology, Tsinghua University, Beijing, China Institute for Artificial Intelligence, Tsinghua University, Beijing, China State Key Lab on Intelligent Technology and Systems, Tsinghua University, Beijing, China 2Search Product Center, WeChat Search Application Department, Tencent, China mooninsiderain@gmail.com Abstract Open relation extraction (OpenRE) aims to extract relational facts from the open-domain corpus. To this end, it discovers relation pat- terns between named entities and then clusters those semantically equivalent patterns into a united relation cluster. Most OpenRE meth- ods typically confine themselves to unsuper- vised paradigms, without taking advantage of existing relational facts in knowledge bases (KBs) and their high-quality labeled instances. To address this issue, we propose Relational Siamese Networks (RSNs) to learn similar- ity metrics of relations from labeled data of pre-defined relations, and then transfer the relational knowledge to identify novel rela- tions in unlabeled data. Experiment results on two real-world datasets show that our frame- work can achieve significant improvements as compared with other state-of-the-art methods. Our code is available at https://github. com/thunlp/RSN. 1 Introduction Relation extraction (RE) aims to extract relational facts between two entities from plain texts. For example, with the sentence “Hayao Miyazaki is the director of the film ‘The Wind Rises’", we can extract a relation “director_of" between two entities “Hayao Miyazaki" and “The Wind Rises". Recent progress in supervised methods to RE has achieved great successes. Supervised meth- ods can effectively learn significant relation se- mantic patterns based on existing labeled data, but the data constructions are time-consuming and human-intensive. To lower the level of super- vision, several semi-supervised approaches have been developed, including bootstrapping, active learning, label propagation (Pawar et al., 2017). ∗ indicates equal contribution † Corresponding author: Z.Liu(liuzy@tsinghua.edu.cn) Relational Siamese Network Novel Relations (Unlabeled) Pre-defined Relations (Auto-labeled/Labeled) 1 … 2 … n … … Testing Instances Novel Relations (Extracted) n+1 … n+2 … n+m … … clustering … every rectangle indicates an unlabeled instance of a novel relation … metric learning every rectangle indicates a pre-defined relation instance relational knowledge transfer Figure 1: A flowchart of our framework. Our model RSN learns from both labeled instances of pre-defined relations and unlabeled instances of new relations, and tries to cluster testing instances of new relations.1 Mintz (2009) also proposes distant supervision to generate training data automatically. It assumes that if two entities have a relation in KBs, all sen- tences that contain these two entities will express this relation. Still, all these approaches can only extract pre-defined relations that have already ap- peared either in human-annotated datasets or KBs. It is hard for them to cover the great variety of novel relational …

Content truncated to 500 words