Learning the Extraction Order of Multiple Relational Facts in a Sentence with Reinforcement Learning

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 367–377, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 367 Learning the Extraction Order of Multiple Relational Facts in a Sentence with Reinforcement Learning Xiangrong Zeng1,2,3, Shizhu He1,2, Daojian Zeng4, Kang Liu1,2, Jun Zhao1,2 1NLPR, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China 2University of Chinese Academy of Sciences, Beijing, 100049, China 3Unisound AI Technology Co.,Ltd, Beijing, 100000, China 4Changsha University of Science & Technology, Changsha, 410114, China {xiangrong.zeng, shizhu.he, kliu, jzhao}@nlpr.ia.ac.cn zengdj@csust.edu.cn Abstract The multiple relation extraction task tries to extract all relational facts from a sentence. Ex- isting works didn’t consider the extraction or- der of relational facts in a sentence. In this pa- per we argue that the extraction order is impor- tant in this task. To take the extraction order into consideration, we apply the reinforcement learning into a sequence-to-sequence model. The proposed model could generate relational facts freely. Widely conducted experiments on two public datasets demonstrate the efficacy of the proposed method. 1 Introduction Relation extraction (RE) is a core task in natural language processing (NLP). RE can be used in in- formation extraction (Wu and Weld, 2010), ques- tion answering (Yih et al., 2015; Dai et al., 2016) and other NLP tasks. Most existing works as- sumed that a sentence only contains one relational facts (a relational fact, or a triplet, contains a rela- tion and two entities). But in fact, a sentence of- ten contains multiple relational facts (Zeng et al., 2018b). The multiple relation extraction task tries to extract all relational facts from a sentence. Existing works on multiple relation extraction task can be divided into five genres. 1) Pseu- doPipeline genre, including Miwa and Bansal (2016); Sun et al. (2018). They first recognized all the entities of the sentence, then extracted features for each entity pair and predicted their relation. They trained the entity recognition model and re- lation prediction model together instead of sepa- rately. Therefore, we call them PseudoPipeline methods. 2) TableFilling genre, including Miwa and Sasaki (2014); Gupta et al. (2016) and Zhang et al. (2017). They maintained a entity-relation table and predicted a semantic tag (either entity tags or relation tags) for each cell in the table. According to the predicted tags, they can recog- nize the entities and the relation between each en- tity pair. 3) NovelTagging genre, including Zheng et al. (2017). This method can be seen as a de- velopment of TableFilling method. They assigned a pre-defined semantic tag to each word of the sentence and collected triplets based on the tags. Their tags include both entity and relation in- formation. Therefore, they don’t need to main- tain a entity-relation table. 4) MultiHeadSelec- tion genre, including Bekoulis et al. (2018a) and Bekoulis et al. (2018b). They first recognized the entities, then they formulated the relation extrac- tion task as a multi-head selection problem. For each entity, …

Content truncated to 500 words