Classification of Depression on social media using Distant Supervision

Abstract

Amidst Covid-19, young adults have experienced major symptoms of anxiety and/or depression disorder (56%). Mental health issues have been spiking all over the world rapidly. People have taken up to social media as a platform to vent about their mental breakdowns. Twitter has seen enormous rise in depressive and anxious tweets in these times, but the downside being that majority of the population has neglected the importance of mental health issues and there are not enough resources to liberate people about it. Also, people hesitate to talk about their mental issues and seek help. So, a machine learning model using distant supervision to detect depression on Twitter is curated. Use of Sentiment140 dataset with 1.6 million records of different tweets. Our training data makes use of Twitter tweets included with emojis, which are classified as noisy labels on a dataset. Further, this paper mentions about how to use models like Support Vector Machine (SVM), Logistic Regression, Naive Bayes, Random Forest, XGBoost to distinguishing tweets between depressive or nondepressive. The purpose behind using multiple models is to achieve highest accuracy when trained with emoticon dataset. The paper’s main contribution is the idea of using tweets with emoticons for distant supervised learning.