Clustering-based Inference for Biomedical Entity Linking

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2598–2608 June 6–11, 2021. ©2021 Association for Computational Linguistics 2598 Clustering-based Inference for Biomedical Entity Linking Rico Angell1, Nicholas Monath1, Sunil Mohan2, Nishant Yadav1, and Andrew McCallum1 1 College of Information and Computer Sciences University of Massachusetts Amherst {rangell,nmonath,nishantyadav,mccallum}@cs.umass.edu 2Chan Zuckerberg Initiative smohan@chanzuckerberg.com Abstract Due to large number of entities in biomedical knowledge bases, only a small fraction of enti- ties have corresponding labelled training data. This necessitates entity linking models which are able to link mentions of unseen entities using learned representations of entities. Pre- vious approaches link each mention indepen- dently, ignoring the relationships within and across documents between the entity mentions. These relations can be very useful for linking mentions in biomedical text where linking de- cisions are often difficult due mentions having a generic or a highly specialized form. In this paper, we introduce a model in which linking decisions can be made not merely by linking to a knowledge base entity but also by group- ing multiple mentions together via clustering and jointly making linking predictions. In ex- periments we improve the state-of-the-art en- tity linking accuracy on two biomedical entity linking datasets including on the largest pub- licly available dataset. 1 Introduction Ambiguity is inherent in the way entities are men- tioned in natural language text. Grounding such am- biguous mentions to their corresponding entities, the task of entity linking, is critical to many applica- tions: automated knowledge base construction and completion (Riedel et al., 2013; Surdeanu et al., 2012), information retrieval (Meij et al., 2014), smart assistants (Balog and Kenter, 2019), ques- tion answering (Dhingra et al., 2020), text mining (Leaman and Lu, 2016; Murty et al., 2018). Consider the excerpt of text from a biomed- ical research paper in Figure 1, the three high- lighted mentions (expression, facial expressions, and facially expressive) all link to the same entity, namely C0517243 - Facial Expresson in the leading biomedical KB, Unified Medical Language System (UMLS) (Bodenreider, 2004). The mention expression is highly ambiguous and easily confused with the more prevalent en- When emotion and expression diverge : The social costs of Parkinson 's disease Patients with Parkinson 's disease are perceived more negatively than their healthy peers , yet it remains unclear what factors contribute to this negative social perception . Based on a cohort of 17 Parkinson 's disease patients and 20 healthy controls , we assessed how naïve raters judge the emotion and emotional intensity displayed in dynamic facial expressions as adults with and without Parkinson 's disease watched emotionally evocative films ( Experiment 1 ) , and how age - matched peers naïve to patients ' disease status judge their social desirability along various dimensions from audiovisual stimuli ( interview excerpts ) recorded after certain films ( Experiment 2 ) . In Experiment 1 , participants with Parkinson 's disease were rated as significantly more facially expressive than healthy controls ... …

Content truncated to 500 words