Classification of Mental Illnesses from Reddit Posts Using Sentence-BERT Embeddings and Neural Networks

The identification and understanding of mental health conditions from social media content forms one of the crucial steps toward enhancing the mental health support systems. This paper attempts to classify mental illnesses by analyzing posts on Reddit by making use of Sentence-BERT (SBERT) embeddings with LSTM as well as BiLSTM networks. SBERT embeddings capture the contextual meaning of user-generated text; on the other hand, BiLSTM networks are inherently good at modeling sequential dependencies and relationships within the text to make much more specific and in-depth classifications of mental health conditions. The SBERT-BiLSTM model achieved strong results, with an accuracy of 70.42% and an F1-score of 0.70, further explaining the importance of advanced neural networks and embeddings toward the detection of mental health concerns from social media data.