Suicidal Ideation Detection Based on Social Media Data in the Context of Bangladesh

Abstract

The increasing use of social media paves the way for solving problems like suicide and urges the need to detect and analyze suicidality. People are suffering from suicidal ideation(SI) across the world. The urgency of early intervention and suicide prevention programs cannot be overstated. Social media platforms are known for self-expression and emotional sharing which offer an opportunity for early detection of SI. Our study aims to use data from social media platforms like Facebook, Twitter, and Reddit. By using machine learning (ML) to the data, we can automatically identify at-risk individuals. We employ two ML based techniques: one topic modeling based deep learning (DL) and one graph neural network (GNN) based models. Our deep learning based model achieves an outstanding accuracy of 87%, while the graph neural network based model achieves a remarkable accuracy of 78%. The deep learning based model surpasses traditional machine learning methods, such as SVM, random forest, and Naive Bayes. Our study may contribute to policy-making approaches which can facilitate early intervention, suicide prevention, and developing a comprehensive decision support system.