Depression Detection Technology Based on Multimodal Social Media Data: A Comprehensive Review and Future Outlook on Text and Image Fusion Analysis

Abstract

Depression is a serious mental disorder, now ranking as the third leading cause of disability worldwide. It profoundly impacts an individual's physical and mental well-being, while also imposing substantial burdens on families, workplaces, and communities. Consequently, early and accurate detection of depression is of paramount importance. Traditional diagnostic methods, which rely on clinical interviews and self-report scales, often fall short due to their subjective nature and reliance on patient cooperation. However, recent advancements in machine learning offer promising new avenues for the objective diagnosis and early intervention of depression. Social media, with its vast troves of user-generated content, serves as a rich database where depressive symptoms can be subtly manifested in users' posts. This paper provides a comprehensive overview of the current methods for detecting depression through social media, focusing on text-image-based multimodal recognition techniques. In addition, the paper identifies existing limitations in these approaches and offers insights into future research directions, aiming to enhance the accuracy and applicability of depression detection technologies. Contribution of This Paper: This work contributes to the field by offering an in-depth analysis of the emerging multimodal methods used for depression detection through social media platforms. It identifies the strengths and weaknesses of current approaches, providing a critical assessment that informs future research. Additionally, the paper expands the scope of traditional depression diagnostics by incorporating advanced machine learning techniques, thereby laying the groundwork for more objective and early intervention strategies.