Affective and Cognitive Distortions-Aided Suicide Risk Prediction for Long-Form Speech in Psychological Support...

Area of Science:

• Artificial Intelligence
• Clinical Psychology
• Speech Processing

Background:

• Speech-based suicide risk prediction is crucial for mental health support.
• Current methods struggle with integrating clinical cues and detecting subtle high-risk signals in speech.

Purpose of the Study:

• To develop an advanced AI model for more accurate suicide risk prediction from speech.
• To enhance the incorporation of clinical insights into speech analysis for mental health.

Main Methods:

• A multi-view feature system combining acoustic, textual, affective, and cognitive distortion cues.
• A hierarchical cascaded decoupling module to manage feature complexity.
• A prior-guided multi-path graph attention network to identify high-risk speech segments and temporal patterns.

Main Results:

• The proposed Affective & Cognitive Distortions-assisted Speech Suicide Risk Prediction Network (ACD-SSRNet) demonstrated superior performance over existing methods.
• Achieved a 2.79% increase in F1-score and a 2.57% increase in accuracy on a real-world hotline dataset.
• Expert evaluation confirmed the model's ability to identify critical affective and cognitive distortion segments linked to suicide risk.

Conclusions:

• ACD-SSRNet offers a significant advancement in AI-driven suicide risk assessment from speech.
• The model's integration of clinical knowledge and sophisticated architecture improves detection accuracy.
• This technology holds promise for enhancing support services in psychological hotlines.