Exploring artificial intelligence in identifying schizophrenic-prone texts from social media

Abstract

Schizophrenia is a detrimental personality disorder that requires prompt diagnosis to prevent destructive behaviors. This study uses AI to analyze the distinctive linguistic characteristics of social media posts by individuals prone to schizophrenia. The research examines Reddit posts associated with schizophrenia employing NLP and ML models such as SVC, Logistic Regression, Naive Bayes, Random Forest, and Decision Tree; Transformer models like DistilBERT and BERT Base; and RNN models like Bi-LSTM and GRU. The DistilBERT model achieves a high accuracy rate of 84%, GRU achieves 79% accuracy, and the logistic regression exhibits an efficiency rate of 83%. With transfer learning, the accuracy of the Distil BERT is 89%, and the GRU model is 83%. The contemporary comparison in the study highlights the importance of consistent data collection for accurate predictive results. The findings have the potential to benefit individuals requiring treatment for recovering from mental illnesses like schizophrenia.