Authors: Mrs. J. Annie Jennifer, Dr. R. Gunasundari
Abstract: The mental health indicators can be found in memes, and it is quite complex since memes consist of both text and images, and one must analyze both elements to understand their meaning. This research proposes a novel deep learning technique named Multi-CNN. Its aim is to detect depression-related signs by analyzing their linguistic and visual content simultaneously in memes . The technology uses both the BERTweet model for natural language processing and ResNet18 features for images from a neural network. It was assessed using a dataset of internet memes annotated according to eight depression indicators. Early stopping, data augmentation, and others helped improve its performance, while results were estimated by means of a weighted F1 score. As the study shows, it is more effective to use linguistic and visual components simultaneously than to employ the model based only on language or solely on image analysis for identifying the presence of depressive signs in memes. The multimodal approach resulted in a weighted F1 score of 0.6846, while the language-based model received 0.6716. Using just the picture is ineffective when it comes to recognizing depression-related memes. The study's findings indicate that visual information and text together create strong cues for investigating mental health issues. Besides, the results point to fresh techniques and technologies that can handle the intricate heterogeneous datasets found in social media.
