Automatic identification of suicide notes with a transformer-based deep learning model.

Area of Science:

• Artificial Intelligence
• Computational Linguistics
• Mental Health Technology

Background:

• Suicide is a major global health concern, with a rise in online suicide notes shared on social media.
• Existing detection methods struggle to capture both local and global semantic nuances in text.
• Accurate detection of online suicide notes is crucial for timely intervention and prevention.

Purpose of the Study:

• To propose an advanced deep learning model for detecting suicide notes online.
• To address the limitations of current methods in capturing comprehensive semantic features.
• To enhance the development of suicide prevention technologies for social media platforms.

Main Methods:

• Development of a transformer-based model named TransformerRNN.
• Integration of a transformer encoder for contextual information extraction.
• Utilization of a Bi-directional Long Short-Term Memory (BiLSTM) structure for long-term dependency analysis.

Main Results:

• TransformerRNN achieved high performance metrics: 95.0% Precision, 94.9% Recall, and 94.9% F1-score.
• The model demonstrated superior performance compared to baseline machine learning and state-of-the-art deep learning approaches.
• Effectively classified suicide notes from a diverse dataset of online posts.

Conclusions:

• The proposed TransformerRNN model is highly effective for classifying suicide notes.
• This technology can significantly contribute to the development of suicide prevention tools.
• Improved detection of online suicide-related content can facilitate timely mental health interventions.