Data-driven Depression Detection System for Textual Data on Twitter using Deep Learning

Mental health is a crucial aspect of human well-being. Social media data like twitter is scrapped and used as the dataset for the depression detection model. This paper proposes to implement a system that will help identify depressive tweets along with depressive user i.e. it is a two-step depression detection process that utilizes deep learning algorithms in conjunction with additional classification for emotion recognition of tweets. This system will include a posts-level system that creates posts representations for all tweets and a user-level system that captures the user representations as whole along with emotion variations and mood variations of a user over a period of time. It takes a data-centric approach which resolves preprocessing issues and class imbalances in the dataset, focusing on providing quality and consistent data for the fine-tuned BERT model which achieves an accuracy of around 97% for the depression detection of tweets.