Deep Learning and Natural Language Processing Techniques for Depression Detection in Social Media Texts: A Comprehensive Review

Abstract

With the increased prevalence of social media, depression is becoming a prominent health issue worldwide and it has opened new opportunities for researchers to explore new techniques for depression detection in social media texts. According to the World Health Organization (WHO) depression may lead to various mental health diseases and suicide if not detected at an early stage. Deep learning and Natural Language Processing are becoming widely adopted techniques among researchers. This review provides a thorough examination of these techniques for depression detection in posts of users. These techniques are harnessed to identify linguistic markers related depression such as sentiment and emotional tone and capture temporal dependencies in texts. Transformer models represent the next level of deep learning techniques, enhanced with self-attention mechanism that enables the automatic analysis of text sequences over time, sematic feature extraction and the interpretation of context-sensitive language. Multimodal approaches using these techniques integrate textual and visual data to improve the accuracy of depression detection. Despite notable advancements, still there are many challenges to address such as data availability, privacy and ethical and model interpretability. Primary aim of this paper is to explore concepts such as evolution of techniques, deep learning and NLP for depression detection, dataset for testing and research gaps and future directions. We conducted a Systematic Literature Review (SLR) on research and review studies published in various conferences and peer-reviewed journals. At last, we provided the brief summary of key findings.