Deep learning-based dimensional emotion recognition for conversational agent-based cognitive behavioral therapy.

Area of Science:

• Artificial Intelligence
• Psychology
• Computational Linguistics

Background:

• Internet-based cognitive behavioral therapy (iCBT) is a scalable and accessible mental health treatment.
• Conversational agents enhance iCBT delivery but often use simplistic categorical emotion models.
• Oversimplified emotion models limit the nuanced understanding required for effective therapy.

Purpose of the Study:

• To develop and evaluate a transformer-based model for dimensional emotion recognition in text.
• To improve the emotional modeling capabilities of conversational agents in iCBT.
• To address the limitations of categorical emotion approaches in digital mental health.

Main Methods:

• Developed a transformer-based model for dimensional emotion recognition (valence, arousal, dominance).
• Fine-tuned the model using a novel dataset of 75,503 dimensional emotion samples.
• Conducted a feasibility study with 20 participants using the model in a conversational agent setting.

Main Results:

• The model achieved high accuracy in recognizing dimensional emotions (Pearson's r = 0.90 for valence, 0.77 for arousal, 0.64 for dominance).
• Significantly outperformed existing state-of-the-art models in dimensional emotion detection.
• Feasibility study confirmed technical effectiveness, usability, acceptance, and empathic understanding.

Conclusions:

• The developed dimensional emotion recognition model represents a significant advancement for conversational AI in iCBT.
• The model enhances personalized and effective therapy experiences by providing deeper emotional insights.
• This work paves the way for more sophisticated and empathetic digital mental health interventions.