CUNI-NU System at Biocreative VIII Track 5: Multi-label Topic Classification of COVID-19 Articles using Dual Attention with SPECTRE
Abstract
Subject-Article classification is an important problem in Scholarly Document Processing to address the huge information overload in the scholarly space. This paper describes the approach of our team CUNI-NU for the Biocreative VII-Track 5 challenge: Litcovid multi-label topic classification for COVID-19 literature [1].
The concerned task aims to automate the manual curation of biomedical articles into seven distinct labels, specifically for the LitCovid data repository. Our best performing model makes use of the SPECTER [2] document embeddings for representing abstract, and titles of scientific articles followed by a Dual-Attention [3] mechanism to perform the multi-label categorization.
We achieve significantly better performance than the baseline methods. We make our code available at https://github.com/Nid989/ CUNI-NU-Biocreative-Track5