Therapy-FIS: A Dataset for Assessing Facilitative Interpersonal Skills in Psychotherapeutic Setting

| Source: arXiv AI

Tags: Therapy-FIS, DyadicFIS, psychotherapy, multimodal datasets, affective computing

Therapy-FIS is described as the first public dataset for automated rating of therapists' facilitative interpersonal skills, built from 890 therapist videos with facial, acoustic, and text signals and a DyadicFIS baseline.

Details

Facilitative Interpersonal Skills assessment rates mental health professionals' interpersonal competence, but relies on costly human observers and has lacked public data. Therapy-FIS contains facial, acoustic, and textual signals from 890 therapist video recordings responding to 8 standardized mock-patient clips, rated on eight FIS subscales plus an overall score. The authors set up two benchmark tasks, therapist-only and joint therapist–mock-patient performance rating, and release a baseline called DyadicFIS using gated fusion and routed readout.