Using hidden Markov modelling to reveal in-session stages in text-based counselling

Area of Science:

• Computational Linguistics
• Psychology
• Human-Computer Interaction

Background:

• Online text-based counseling platforms are increasingly prevalent.
• Understanding the dynamics of successful counseling sessions is crucial for service improvement.
• Previous analyses often lack quantitative methods to identify distinct session stages.

Purpose of the Study:

• To apply Hidden Markov Models (HMMs) to analyze text-based counseling session transcripts.
• To identify latent stages within counseling sessions based on word distributions.
• To compare the progression patterns of more versus less satisfying counseling sessions.

Main Methods:

• Utilized transcripts from 2589 online text-based counseling sessions.
• Employed a message-level Hidden Markov Model (HMM) to infer session stages.
• Categorized sessions into 'more satisfying' and 'less satisfying' based on post-session surveys.

Main Results:

• Identified five distinct stages: Rapport-building, Problem-identification, Problem-exploration, Problem-solving, and Wrap-up.
• More satisfying sessions demonstrated significantly more efficient rapport building (7.5% duration) and problem exploration (28.5%).
• Highly satisfying sessions featured more elaborated solution development (46.6%) and concise conclusions (8.2%).

Conclusions:

• Text-based counseling progresses through identifiable stages.
• Efficient initial engagement, thorough exploration, and focused problem-solving correlate with higher session satisfaction.
• HMMs provide a robust method for analyzing and optimizing digital therapeutic interactions.