This research demonstrates how preference learning can significantly improve LLMs' ability to deliver effective psycho-counseling responses, helping bridge the gap between mental health needs and available support.

• Created PsychoTherapyPref dataset with 15,000+ examples of therapeutic responses
• Developed MindGuide, a model fine-tuned with preference optimization
• Achieved superior performance compared to both general LLMs and specialized mental health models
• Identified key attributes of high-quality therapeutic responses

This advancement matters because it addresses critical shortages in mental healthcare access without compromising patient privacy, potentially enabling more accessible, consistent support for those in need.

Preference Learning Unlocks LLMs' Psycho-Counseling Skills