JMIR Publications

ABSTRACT

In an era dominated by social media conversations, it's pivotal to comprehend how suicide, a critical public health issue, is discussed online. In this research, we draw upon established psychological literature to identify core topics linked to suicide, such as mental health challenges, relationship conflicts, and financial distress. Then, we undertake a comprehensive analysis of suicide-related data sourced from social media, employing a guided topic modelling technique to extract the most frequently discussed subjects based on the risk factors from the literature. Our study shows that many critical suicide-related topics, such as those related to racism and marginalized communities, are underrepresented in this data. To address this issue, we explore the possibility of generating topic-diverse synthetic data using generative models, such as ChatGPT, for data augmentation. We comprehensively evaluated and analyzed the synthetic dataset to assess its readability, complexity, topic diversity, and utility in training classifiers compared to real datasets. Our results demonstrate that such datasets can be useful to forge a more enriched understanding of online suicide discussions as well as build more accurate machine learning models for suicidal narrative detection on social media.