Multi-label Multi-task Deep Learning for Behavioral Coding.

Area of Science:

• Computational Linguistics
• Psychology
• Machine Learning

Background:

• Behavioral coding in psychotherapy involves annotating human interaction data with labels.
• Understanding therapist-client interactions is crucial for improving mental healthcare.
• Existing methods may not fully capture the complexity of behaviors in therapy sessions.

Purpose of the Study:

• To propose and evaluate a methodology for estimating human behaviors in psychotherapy using multi-label and multi-task learning.
• To compare the effectiveness of different learning paradigms (single vs. multiple label/task) for behavioral estimation.
• To assess the impact of incorporating turn context on prediction performance.

Main Methods:

• Development of a methodology based on multi-label and multi-task learning paradigms.
• Utilizing two distinct corpora of therapist-client interactions from psychotherapy sessions.
• Experimental comparison of single-label, multi-label, single-task, and multi-task learning approaches.
• Evaluation of performance with and without the inclusion of turn context.

Main Results:

• The proposed multi-label and multi-task learning approaches demonstrate significant prediction performance gains.
• Incorporating turn context further enhances the accuracy of behavior estimation.
• The models provide valuable insights into complex human interactions within psychotherapy.

Conclusions:

• Multi-label and multi-task learning represent effective paradigms for estimating human behaviors in psychotherapy.
• Turn context is an important feature for improving the accuracy of behavioral analysis.
• These computational methods offer a promising avenue for deeper understanding and analysis of therapeutic dialogues.