Back

Fine-tuning Large Language Models in Behavioral Psychology for Scalable Physical Activity Coaching

Mantena, S. D.; Johnson, A.; Oppezzo, M.; Schuetz, N.; Tolas, A.; Doijad, R.; Mattsson, C. M.; Lawrie, A.; Ramirez-Posada, M.; Linos, E.; King, A. C.; Rodriguez, F.; Kim, D. S.; Ashley, E.

2025-02-21 cardiovascular medicine
10.1101/2025.02.19.25322559 medRxiv
Show abstract

Personalized, smartphone-based coaching improves physical activity but relies on static, human-crafted messages. We introduce My Heart Counts (MHC)-Coach, a large language model fine-tuned on the Transtheoretical Model of Change. MHC-Coach generates messages tailored to an individuals psychology (their "stage of change"), providing personalized support to foster long-term physical activity behavior change. To evaluate MHC-Coachs efficacy, 632 participants compared human-expert and MHC-Coach text-based interventions encouraging physical activity. Among messages matched to an individuals stage of change, 68.0% (N=430) preferred MHC-Coach-generated messages (P < 0.001). Blinded behavioral science experts (N=2) rated MHC-Coach messages higher than human-expert messages for perceived effectiveness (4.4 vs. 2.8) and Transtheoretical Model alignment (4.1 vs. 3.5) on a 5-point Likert scale. This work demonstrates how language models can operationalize behavioral science frameworks for personalized health coaching, promoting long-term physical activity and potentially reducing cardiovascular disease risk at scale.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.