DETECTING DEPRESSION IN SOCIAL MEDIA USERS USING ADVANCED DEEP LEARNING APPROACHES

Abstract

Due to stigma and a lack of easily available diagnostic resources, depression is a common mental health illness that frequently goes undiagnosed. As social media has grown in popularity, people are expressing their feelings and psychological states more and more online, which offer a wealth of information for mental health research. In order to detect depression from multimodal data—specifically, textual content from social media and user behavioral features—this thesis proposes a hybrid deep learning model that combines Bidirectional Encoder Representations from Transformers (BERT), Bidirectional Long Short-Term Memory (BiLSTM), and Convolutional Neural Networks (CNN). The design makes use of CNN's prowess in identifying local patterns in text embedding, BiLSTM's capacity to model sequential relationships, and BERT's contextual language understanding to capture subtle semantic representations. To enhance the model's input space, behavioral data like posting frequency, activity time, and interaction patterns are combined with textual features. A carefully selected multimodal dataset was used to train and assess the hybrid model, which was then contrasted with a number of baseline models, such as deep learning variations and conventional machine learning classifiers. In terms of accuracy, precision, recall, and AUC, experimental results show that the suggested hybrid model performs noticeably better than baseline methods, underscoring its efficacy and resilience in identifying depressed symptoms. This thesis offers a solid basis for further study in the area of intelligent psychological diagnosis and highlights the possibilities of integrating linguistic and behavioral modalities for automated mental health assessment.