Extracting a Knowledge Base of Mechanisms from COVID-19 Papers

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4489–4503 June 6–11, 2021. ©2021 Association for Computational Linguistics 4489 Extracting a Knowledge Base of Mechanisms from COVID-19 Papers Tom Hope♣,♠∗ Aida Amini ♣* David Wadden♣ Madeleine van Zuylen♠ Sravanthi Parasa♡ Eric Horvitz♢ Daniel Weld♣,♠ Roy Schwartz♥ Hannaneh Hajishirzi♣,♠ ♣ Paul G. Allen School for Computer Science & Engineering, University of Washington ♠ Allen Institute for Artificial Intelligence ♡ Swedish Medical Group ♢ Microsoft Research ♥ The Hebrew University of Jerusalem, Israel {tomh,aidaa,danw, hannah}@allenai.org Abstract The COVID-19 pandemic has spawned a di- verse body of scientific literature that is chal- lenging to navigate, stimulating interest in automated tools to help find useful knowl- edge. We pursue the construction of a knowl- edge base (KB) of mechanisms—a fundamen- tal concept across the sciences, which encom- passes activities, functions and causal rela- tions, ranging from cellular processes to eco- nomic impacts. We extract this information from the natural language of scientific pa- pers by developing a broad, unified schema that strikes a balance between relevance and breadth. We annotate a dataset of mechanisms with our schema and train a model to extract mechanism relations from papers. Our exper- iments demonstrate the utility of our KB in supporting interdisciplinary scientific search over COVID-19 literature, outperforming the prominent PubMed search in a study with clin- ical experts. Our search engine, dataset and code are publicly available.1 1 Introduction “Some experts are familiar with one field, such as AI or nanotechnology [...] no one is capable of connecting the dots and seeing how breakthroughs in AI might impact nanotechnology, or vice versa.” –Yuval Noah Harari, Homo Deus, 2016 The effort to mitigate the COVID-19 pandemic is an interdisciplinary endeavor the world has rarely seen (Apuzzo and Kirkpatrick, 2020). As one re- cent example, expertise in virology, physics, epi- demiology and engineering enabled a group of 200 scientists to understand and bring attention to the ∗*Equal contribution. 1https://covidmechanisms.apps.allenai.org/ … a deep learning framework for design of antiviral candidate drugs Temperature increase can facilitate the destruction of SARS-COV-2 gpl16 antiserum blocks binding of virions to cellular receptors ...food price inflation is an unintended consequence of COVID-19 containment measures Retrieved from CORD-19 papers Ent1: deep learning Ent2: drugs Query mechanism relations Ent1: heat Ent2: SARS-CoV-2 Ent1: ? Ent2: block viral binding Ent1: COVID-19 Ent2: prices COMB: Covid Mechanism KB Relation? Ent1 Ent2 (1) (2) Figure 1: Our COVID-19 Mechanism KB (COMB) is extracted from scientific papers and can be searched for diverse activities, functions and influences (1), retriev- ing relations from the literature (2). airborne transmissibility of the SARS-CoV-2 virus (Morawska et al., 2020). The diverse and rapidly expanding body of past and present findings related to COVID-19 (Wang et al., 2020b) makes it chal- lenging to keep up, hindering scientists’ pace in making new discoveries and connections. Research in natural language processing (NLP) has provided important resources to extract fine- grained relations from scientific papers in specific …

Content truncated to 500 words