Comprehensive Summary
This study, published in the Journal of Medical Internet Research, investigates how young people’s emotional expression changes during real-time, in 6207 randomly extracted text-based counseling sessions. Researchers analyzed over 6,000 first-time counseling sessions from “Open Up,” a 24/7 online chat service in Hong Kong, using GPT-4 AI tool to classify each message by positive, neutral, or negative sentiment and then modeling how these sentiments shifted across five equal stages of each session. Three distinct trajectories were identified: a steady improvement pattern (18.9% of sessions), where users became increasingly positive; a deterioration pattern (18%), where emotions declined over time; and the most common, a dip-then-rebound pattern (63.1%), where initial negativity gave way to stabilization and eventual improvement. Factors such as suicidal ideation, family or health-related problems, anonymity, and early session termination were linked to the deterioration pathway. The authors emphasize that sentiment does not follow a uniform trajectory and that monitoring these changes in real time could allow counselors to tailor their approaches dynamically.
Outcomes and Implications
This research highlights how AI can track emotional shifts in mental health interventions, helping to identify when clients may be at risk of worsening distress. Clinically, it suggests that integrating sentiment-monitoring tools into digital counseling platforms could provide early warning signals, enabling counselors to intervene more effectively with high-risk users, particularly those expressing suicidal thoughts. While the study was limited to a specific population and context, its framework could inform future clinical tools for broader telehealth and online mental health services. The authors propose that real-time sentiment analysis could be implemented as part of ongoing digital triage systems, potentially becoming clinically relevant in the near future as AI systems are more deeply embedded into mental health care workflows.