ChemNER: Fine-Grained Chemistry Named Entity Recognition with Ontology-Guided Distant Supervision

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 5227–5240 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 5227 CHEMNER: Fine-Grained Chemistry Named Entity Recognition with Ontology-Guided Distant Supervision Xuan Wang1∗, Vivian Hu1∗, Xiangchen Song2, Shweta Garg1, Jinfeng Xiao1 and Jiawei Han1 1University of Illinois at Urbana-Champaign, IL, USA 2Carnegie Mellon University, PA, USA 1{xwang174,vivianh2,shwetag2,jxiao13,hanj}@illinois.edu 2xiangchensong@cmu.edu Abstract Scientific literature analysis needs fine-grained named entity recognition (NER) to provide a wide range of information for scientific discovery. For example, chemistry research needs to study dozens to hundreds of dis- tinct, fine-grained entity types, making con- sistent and accurate annotation difficult even for crowds of domain experts. On the other hand, domain-specific ontologies and knowl- edge bases (KBs) can be easily accessed, con- structed, or integrated, which makes distant supervision realistic for fine-grained chem- istry NER. In distant supervision, training la- bels are generated by matching mentions in a document with the concepts in the knowl- edge bases (KBs). However, this kind of KB-matching suffers from two major chal- lenges: incomplete annotation and noisy anno- tation. We propose CHEMNER, an ontology- guided, distantly-supervised method for fine- grained chemistry NER to tackle these chal- lenges. It leverages the chemistry type ontol- ogy structure to generate distant labels with novel methods of flexible KB-matching and ontology-guided multi-type disambiguation. It significantly improves the distant label gen- eration for the subsequent sequence labeling model training. We also provide an expert- labeled, chemistry NER dataset with 62 fine- grained chemistry types (e.g., chemical com- pounds and chemical reactions). Experimen- tal results show that CHEMNER is highly ef- fective, outperforming substantially the state- of-the-art NER methods (with .25 absolute F1 score improvement). 1 Introduction Named entity recognition (NER) is a fundamen- tal step in scientific literature analysis to build AI-driven systems for molecular discovery, syn- thetic strategy designing, and manufacturing (Xie ∗The first two authors contributed equally to this work and should be considered as joint first authors. et al., 2013; Szklarczyk et al., 2015; Huang et al., 2015; Szklarczyk et al., 2017; de Almeida et al., 2019). It aims to locate and classify entity mentions (e.g., “Suzuki-Miyaura cross-coupling reactions”) from unstructured text into pre-defined categories (e.g., “coupling reactions”). In the chemistry do- main, previous NER studies are mostly focused on one coarse-grained entity type (i.e., chemicals) (Krallinger et al., 2015; He et al., 2020; Watanabe et al., 2019) and rely on large amounts of manually- annotated data for training deep learning models (Chiu and Nichols, 2016; Ma and Hovy, 2016; Lam- ple et al., 2016; Wang et al., 2019b; Devlin et al., 2019; Liu et al., 2019). In real-world applications, it is important to recognize chemistry entities on diverse and fine- grained types (e.g., “inorganic phophorus com- pounds”, “coupling reactions” and “catalysts”) to provide a wide range of information for scientific discovery. It will need dozens to hundreds of dis- tinct types, making consistent and accurate anno- tation difficult even for domain experts. On the other hand, the domain-specific ontologies and knowledge bases …

Content truncated to 500 words