Zero-shot Medical Entity Retrieval without Annotation: Learning From Rich Knowledge Graph Semantics

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 2401–2405 August 1–6, 2021. ©2021 Association for Computational Linguistics 2401 Zero-shot Medical Entity Retrieval without Annotation: Learning From Rich Knowledge Graph Semantics Luyang Kong Amazon AI luyankon@amazon.com Christopher Winestock Amazon AI winestoc@amazon.com Parminder Bhatia Amazon AI parmib@amazon.com Abstract Medical entity retrieval is an integral compo- nent for understanding and communicating in- formation across various health systems. Cur- rent approaches tend to work well on specific medical domains but generalize poorly to un- seen sub-specialties. This is of increasing con- cern under a public health crisis as new med- ical conditions and drug treatments come to light frequently. Zero-shot retrieval is chal- lenging due to the high degree of ambiguity and variability in medical corpora, making it difficult to build an accurate similarity mea- sure between mentions and concepts. Medi- cal knowledge graphs (KG), however, contain rich semantics including large numbers of syn- onyms as well as its curated graphical struc- tures. To take advantage of this valuable infor- mation, we propose a suite of learning tasks de- signed for training efficient zero-shot entity re- trieval models. Without requiring any human annotation, our knowledge graph enriched ar- chitecture significantly outperforms common zero-shot benchmarks including BM25 and Clinical BERT with 7% to 30% higher re- call across multiple major medical ontologies, such as UMLS, SNOMED and ICD-10. 1 Introduction Entity retrieval is the task of linking mentions of named entities to concepts in a curated knowledge graph (KG). This allows medical researchers and clinicians to search medical literature easily using standardized codes and terms to improve patient care. Training an effective entity retrieval system often requires high quality annotations, which are expensive and slow to produce in the medical do- main. It is therefore not feasible to annotate enough data to cover the millions of concepts in a medical KG, and difficult to adapt quickly enough to those newly appeared medical conditions and drug treat- ments under a public health crisis. Hence, a robust medical entity retrieval system is expected to have decent performance in a zero-shot scenario. Zero-shot retrieval is challenging due to the com- plexity of medical corpora - large numbers of am- biguous terms, copious acronyms and synonymous terms. It is difficult to build an accurate similar- ity measure which can detect the true relatedness between a mention and a concept even when their surface forms differ greatly. Early entity retrieval systems use string matching methods such as exact match, approximate match (Hanisch et al., 2005) and weighted keyword match e.g. BM25 (Wang et al., 2020). Although anno- tated training data is not required, such systems typically lack the ability to handle synonyms and paraphrases with large surface form differences. In recent years, large scale pretraining (Devlin et al., 2019) has been widely adopted in the medical do- main such as Clinical BERT (Alsentzer et al., 2019) and BioBERT (Lee et al., 2019). Agarwal et al. (2019) also integrates graph structure information during pretraining. Most of them, however, require …

Content truncated to 500 words