Easy First Relation Extraction with Information Redundancy

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 3851–3861, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 3851 Easy First Relation Extraction with Information Redundancy Shuai Ma1,2 Gang Wang1,2 Yansong Feng3 Jinpeng Huai1,2 1SKLSDE Lab, Beihang University, China 2 Beijing Advanced Innovation Center for Big Data and Brain Computing, Beijing, China 3Peking University, China {mashuai, iegwang, huaijp}@buaa.edu.cn fengyansong@pku.edu.cn Abstract Many existing relation extraction (RE) models make decisions globally using integer linear programming (ILP). However, it is nontrivial to make use of integer linear programming as a blackbox solver for RE. Its cost of time and memory may become unacceptable with the increase of data scale, and redundant informa- tion needs to be encoded cautiously for ILP. In this paper, we propose an easy first approach for relation extraction with information redun- dancies, embedded in the results produced by local sentence level extractors, during which conflict decisions are resolved with domain and uniqueness constraints. Information re- dundancies are leveraged to support both easy first collective inference for easy decisions in the first stage and ILP for hard decisions in a subsequent stage. Experimental study shows that our approach improves the efficiency and accuracy of RE, and outperforms both ILP and neural network-based methods. 1 Introduction Relation extraction (RE) has been extensively studied due to its crucial role for many knowl- edge based applications (Fan et al., 2012; Zhang et al., 2012; Chen et al., 2014a), such as ques- tion answering and knowledge graph. There are two types of relation extractors: local and global. The former identify relationships between en- tity pairs individually according to the local fea- tures of sentences, e.g., lexical and syntactic fea- tures (Bunescu and Mooney, 2007; Mintz et al., 2009; Surdeanu et al., 2010; Hoffmann et al., 2011; Surdeanu et al., 2012; Zeng et al., 2015; Lin et al., 2016). The latter make decisions by further considering joint features of the entire corpus (Yao et al., 2010; Li et al., 2011, 2013; de Lacalle and Lapata, 2013; Chen et al., 2014a). Global rela- tion extractors are able to resolve conflict deci- sions, and to utilize the dependencies among ex- tracted facts to improve the performance (Li et al., 2011; Chen et al., 2014a), commonly by formal- izing RE as a constrained optimization problem and solving RE with integer linear programming (ILP) (Roth and Yih, 2004; Choi et al., 2006; Roth and Yih, 2007; Li et al., 2011; Ji et al., 2013; Chen et al., 2014a,b). However, global optimiza- tion remains dominated by the ILP solvers, suffer- ing from heavy time expenses. Using integer linear programming for RE as a blackbox solver is a challenging task. First, with the increase of entity pairs and candidate relations, the variables and constraints encoded for ILP in- crease dramatically, which in return may consume too much computing time and memory. Second, redundant information needs to be encoded cau- tiously. Consider <United States, capital …

Content truncated to 500 words