Leveraging Dependency Forest for Neural Medical Relation Extraction

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 208–218, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 208 Leveraging Dependency Forest for Neural Medical Relation Extraction Linfeng Song1,2, Yue Zhang3,4⇤, Daniel Gildea1, Mo Yu5, Zhiguo Wang6 and Jinsong Su7 1University of Rochester, Rochester, NY, USA 2Tencent AI Lab, Bellevue, WA, USA 3Institute of Advanced Technology, Westlake Institute for Advanced Study 4School of Engineering, Westlake Universtiy 5IBM T.J. Watson Research, Yorktown Heights, NY, USA 6Amazon AWS, New York, NY, USA 7Xiamen University, Xiamen, China Abstract Medical relation extraction discovers relations between entity mentions in text, such as re- search articles. For this task, dependency syn- tax has been recognized as a crucial source of features. Yet in the medical domain, 1- best parse trees suffer from relatively low ac- curacies, diminishing their usefulness. We investigate a method to alleviate this prob- lem by utilizing dependency forests. Forests contain many possible decisions and therefore have higher recall but more noise compared with 1-best outputs. A graph neural network is used to represent the forests, automatically distinguishing the useful syntactic information from parsing noise. Results on two biomedi- cal benchmarks show that our method outper- forms the standard tree-based methods, giving the state-of-the-art results in the literature. 1 Introduction The sheer amount of medical articles and their rapid growth prevent researchers from receiv- ing comprehensive literature knowledge by direct reading. This can hamper both medical research and clinical diagnosis. NLP techniques have been used for automating the knowledge extraction pro- cess from the medical literature (Friedman et al., 2001; Yu and Agichtein, 2003; Hirschman et al., 2005; Xu et al., 2010; Sondhi et al., 2010; Abacha and Zweigenbaum, 2011). Along this line of work, a long-standing task is relation extraction, which mines factual knowledge from free text by labeling relations between entity mentions. As shown in Figure 1, the sub-clause “previously observed cy- tochrome P450 3A4 ( CYP3A4 ) interaction of the dual orexin receptor antagonist almorexant” con- tains two entities, namely “orexin receptor” and “almorexant”. There is an “adversary” relation be- tween these two entities, denoted as“CPR:6”. ⇤Yue Zhang is the corresponding author ... observed ... interaction of orexin receptor antagonist almorexant amod nmod comp comp comp (a) ... observed ... interaction of orexin receptor antagonist almorexant amod nmod comp comp comp nmod comp (b) Figure 1: (a) 1-best dependency tree and (b) depen- dency forest for a medical-domain sentence, where edge label “comp” represents “compound”. Associated mentions are in different colors. Some irrelevant words and edges are omitted for simplicity. Previous work has shown that dependency syntax is important for guiding relation extrac- tion (Culotta and Sorensen, 2004; Bunescu and Mooney, 2005; Liu et al., 2015; Gormley et al., 2015; Xu et al., 2015a,b; Miwa and Bansal, 2016; Zhang et al., 2018b), especially in biological and medical domains (Quirk and Poon, 2017; Peng et al., 2017; Song et al., 2018b). Compared with sequential …

Content truncated to 500 words