Assessing Heterogeneity in Sentiment Changes in Text-Based Counseling: Latent Class Trajectory Analysis.

Area of Science:

• Digital Mental Health
• Computational Social Science
• Clinical Psychology

Background:

• Online text-based counseling is growing, but its nature presents challenges in monitoring user emotional shifts.
• Anonymity and text format hinder personalized interventions, potentially impacting service effectiveness and user satisfaction.

Purpose of the Study:

• To identify distinct within-session sentiment trajectories of help-seekers in online text-based counseling.
• To examine key variables associated with different sentiment trajectory memberships.

Main Methods:

• Latent class trajectory analysis using a growth mixture model (GMM) on 6207 counseling sessions.
• Sentiment scores of help-seeker messages, labeled by ChatGPT, were used for trajectory modeling.
• Multinomial logistic regression identified variables associated with class membership.

Main Results:

• Three distinct sentiment trajectories were identified: steady improvement (18.9%), deterioration (18.0%), and dip-then-rebound (63.1%).
• The deterioration trajectory was significantly associated with suicidal ideation, family/physical health concerns, and premature session departure.

Conclusions:

• Distinct within-session sentiment trajectories exist in online text-based counseling.
• Early identification of these trajectories can enable counselors to adapt interventions, improving effectiveness and satisfaction in digital mental health services.