Deep learning techniques for suicide and depression detection from online social media: A scoping review

Psychological health, i.e., citizens’ emotional and mental well-being, is one of the most neglected public health issues. Depression is the most common mental health issue and the leading cause of suicide and self-injurious behavior. Clinical diagnosis of these mental health issues is expensive and also ignored due to social stigma and lack of awareness. Nowadays, online social media has become the preferred mode of communication, which people use to express their thoughts, feelings, and emotions. Hence, user-generated content from online social media can be leveraged for non-clinical mental health assessment and surveillance. Conventional machine learning and NLP techniques have been used for the automated detection of mental health issues using social media data for a very long time now. However, the objective of our research is to study the applications of deep learning techniques for early detection and non-clinical, predictive diagnosis of depression, self-harm, and suicide ideation from online social network content only. To the best of our knowledge, we did not find any systematic literature review that studies the applications of deep learning techniques in this domain. In order to address this research gap, we conducted a systematic literature review (SLR) of 96 relevant research studies published until date that have applied deep learning techniques for detecting depression and suicide or self-harm behavior from social media content. Our work comprehensively covers state-of-the-art w.r.t. techniques, features, datasets, and performance metrics, which can be of great value to researchers. We enumerate all the available datasets in this domain and discuss their characteristics. We also discuss the research gaps, challenges, and future research directions for advancing & catalyzing research in this domain. To the best of our knowledge, our study is the only and the most recent survey for this domain covering the latest research published until date. Based on our learnings from this review, we have also proposed a framework for mental health surveillance. We believe the findings of our work will be beneficial for researchers working in this domain.