MatchPrompt: Prompt-based Open Relation Extraction with Semantic Consistency Guided Clustering

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 7875 - 7888 December 7-11, 2022 ©2022 Association for Computational Linguistics MatchPrompt: Prompt-based Open Relation Extraction with Semantic Consistency Guided Clustering Jiaxin Wang1,2 , Lingling Zhang1,2∗, Jun Liu1,2 , Xi Liang1,2 , Yujie Zhong1,2 , Yaqiang Wu2,3, 1Shaanxi Provincial Key Laboratory of Big Data Knowledge Engineering, School of Computer Science and Technology, Xi’an Jiaotong University, China 2National Engineering Lab for Big Data Analytics, Xi’an Jiaotong University, China 3Lenovo Research, Beijing, China jiaxinwangg@outlook.com, {zhanglling,liukeen}@xjtu.edu.cn {lliangxii,yjzhong6}@gmail.com, wuyqe@lenovo.com Abstract Relation clustering is a general approach for open relation extraction (OpenRE). Current methods have two major problems. One is that their good performance relies on large amounts of labeled and pre-defined relational instances for pre-training, which are costly to acquire in reality. The other is that they only focus on learning a high-dimensional metric space to measure the similarity of novel rela- tions and ignore the specific relational repre- sentations of clusters. In this work, we propose a new prompt-based framework named Match- Prompt, which can realize OpenRE with effi- cient knowledge transfer from only a few pre- defined relational instances as well as mine the specific meanings for cluster interpretability. To our best knowledge, we are the first to intro- duce a prompt-based framework for unlabeled clustering. Experimental results on different datasets show that MatchPrompt achieves the new SOTA results for OpenRE. 1 Introduction Relation Extraction (RE) is one of the most es- sential technology for knowledge graph construc- tion (Church and Bian, 2021; Mirtaheri, 2021). It aims to detect the relation between two entities in a sentence. For example, when given the en- tity pair (San Diego, the United States) in the sen- tence San Diego is located in the United States, a relation extraction model should predict the re- lation located in between these two entities. De- spite traditional RE models having achieved great success (Li et al., 2021; Ye et al., 2022), they are designed based on pre-defined relations and are incapable of extracting novel relations in the real world. Therefore, Open Relation Extraction (OpenRE) has been proposed to extract relations without pre-defined types from the open-domain corpus. Methods for OpenRE can be divided into ∗ Corresponding author Yang Yan [MASK] Her sister Yang Yan was Emperor Wu's wife (a) Prompt-based Traditional RE Strategy (b) Prompt-based OpenRE Strategy [MASK] Emma has been married to Polanski for years. MLM Head Emma Emperor Wu Polanski Labeled Sentences Unlabeled Sentences Template Template spouse relation type spouse relation type label words married partner ... label words married partner ... [MASK] Her sister Yang Yan was Emperor Wu's wife Yang Yan Emperor Wu Template Unlabeled Sentences [MASK] Her sister Yang Yan was Emperor Wu's wife Yang Yan Emperor Wu Template Unlabeled Sentences Match MLM Head MLM Head Match MLM Head MLM Head Figure 1: Comparison of prompt-based traditional RE strategy and OpenRE strategy. two categories based on the form of extracted novel relations. The former is called open information extraction (OIE) (Yates …

Content truncated to 500 words