Depression Detection from Meme Image Posts Transformer-Based Text Analysis

In the most recent times memes have experienced high levels of popularity as a method of elf-expression and it is particularly the younger users who use social media. Despite their common comedy, a lot of memes covertly represent psychiatrical disorders e.g. depressive disorders. Based on just retrieved text, this paper proposes a transformer-based Natural Language Processing (NLP) method to detect depressed content in memes. Memes are optimized with DistilBERT model for binary classification using OCR-derived meme captions classified into depressive and non-depressive categories. The model’s strong overall performance brings up the accuracy of 86.5% and F1-score equal to 0.87 for the” Depressed” class after 20 training epochs that show efficiency of text based analysis in determining mental states even without presence of visual content. By doing so, this approach fills the gap between passively monitoring mental health through social media content and buoying a more scalable, non-invasive technique for detecting mental health signals among young demographics towards digital mental health assessment conversation.