Natural Language Transfer Learning for Physiological Textual Similarity

Abstract

Understanding textual and language information has always been one of the primary research concerns of artificial intelligence, as the crucial function it plays in communication. The biomedical domain has experienced a surge in the availability of data in the form of text. This collection of information has opened avenues to a plethora of automated applications. In this work, the nascent technique of Natural Language Transfer Learning is employed for Physiological Computing. This methodology measures the semantic similarity between medical text utilising pre-trained language models such as BERT and RoBERTa. Using the proposed methodology 90% accuracy over the BioSSES dataset has been obtained. Henceforth, transfer learning proves to be an effectual strategy for NLP tasks that belong to varied fields.