Applying Multimodal Deep Learning for Identifying Mental Health Indicators in Instagram Content

This study explores the application of multimodal deep learning to identify mental health indicators within Instagram content. Leveraging the unique combination of images, captions, and hashtags on Instagram, the research addresses a significant gap in current mental health detection methods that typically rely on single data modalities. A deep learning model was developed to integrate visual and textual features through advanced multimodal fusion techniques. The model was trained and evaluated on a dataset of Instagram posts, achieving an overall accuracy of 69%. The results highlight the model's effectiveness in identifying posts related to mental health issues, particularly those that are “Teridentifikasi.” However, the model also revealed challenges in precisely identifying “Tidak Teridentifikasi” posts, indicating areas for future refinement. This research contributes to the field by advancing the understanding of multimodal content in social media and its implications for mental health detection, offering a more comprehensive and nuanced approach compared to traditional methods. The findings have the potential to inform future developments in computational social science and mental health assessment tools.