SSN@LT-EDI-ACL2022: Transfer Learning using BERT for Detecting Signs of Depression from Social Media Texts

Abstract

Depression is one of the most common mental issues faced by people. Detecting signs of depression early on can help in the treatment and prevention of extreme outcomes like suicide. Since the advent of the internet, people have felt more comfortable discussing topics like depression online due to the anonymity it provides. This shared task has used data scraped from various social media sites and aims to develop models that detect signs and the severity of depression effectively. In this paper, we employ transfer learning by applying enhanced BERT model trained for Wikipedia dataset to the social media text and perform text classification. The model gives a F1-score of 63.8% which was reasonably better than the other competing models.