Leveraging multi-class sentiment analysis on social media text for detecting signs of depression

Abstract

Depression is a major disorder that causes an insistent feeling of sadness which affects the thinking and behaviour of the person that leads to variety of emotional and physical problems. Early detection of depression is most important since it is associated with sui-cide and it is also the main cause for somatic diseases. In recent days, people are freely conveying what they think directly through any of the social media. In the proposed re-search work, machine learning, deep learning and hybrid learning approaches are utilized to perform sentiment analysis in order to determine a person's indicators of sadness from their social media postings. All of the aforementioned models use English language social media postings from the shared task introduced by ACL 2022 and categories depression signs into three groups such as not depressed, moderately depressed and severely de-pressed. The experimental results show that among all the machine learning, deep learning and hybrid learning techniques, CNN-LSTM model obtained highest precision of 0.93 for moderate depressed and severe depression categories.