From Reddit to Reality: Identifying Schizophrenia Through a BERT-BiLSTM Analysis of Online Communities

Schizophrenia is a chronic mental condition that disrupts thoughts, behaviors, and emotions, often manifesting during late adolescence or early adulthood. In this study, we propose a hybrid Natural Language Processing (NLP) approach that integrates Bidirectional Encoder Representations from Transformers (BERT) for feature extraction and Bidirectional Long Short-Term Memory (BiLSTM) networks for classification. Our goal is to distinguish between healthy individuals and schizophrenia-affected users based on their Reddit posts. Unlike prior research relying solely on traditional machine learning techniques, we evaluate and compare standalone BERT, standalone BiLSTM, and a novel BERT-BiLSTM hybrid architecture. We further investigate the limitations of standalone models and demonstrate how the hybrid model improves performance. The Reddit Mental Health Dataset serves as the data source for this study, ensuring a rich and diverse sample. Our findings highlight the potential of combining deep contextual understanding with sequential dependency learning for effective schizophrenia detection from online communities. Reddit provides a research database, which is freely accessible without the permission or consent of forum posters, therefore, the Reddit database is used here.Problem Statement: Identifying schizophrenia from user-generated Reddit posts remains challenging due to linguistic complexity and lack of clinical supervision. This study addresses the need for robust NLP models that can effectively differentiate between schizophrenia-affected and healthy users based on language features.