Detection of Psychological States in Persian Social Media Posts Using Advanced Deep Learning Architectures with Attention Mechanisms

In recent years, social media platforms have emerged as significant spaces for users to express their psychological states. Early detection of mental health conditions such as depression, anxiety, and suicidal ideation in these environments can play a vital role in timely intervention and prevention. This study presents a system based on advanced deep learning architectures incorporating attention mechanisms for the automatic detection of psychological states in Persian-speaking users. Data were collected from platforms such as Instagram, Twitter, Telegram, and specialized psychiatric websites, and were labeled by a team of expert psychiatrists. After comprehensive preprocessing of Persian texts, several models including LSTM, BiLSTM, and Bidirectional GRU with Attention were evaluated. Results indicate that the Bidirectional GRU with Attention achieved the highest performance with an accuracy of 0.87. By focusing on Persian language data and utilizing expert-labeled realworld content, this research contributes a significant step toward the development of intelligent tools for mental health monitoring in digital environments.