Designing a just-in-time adaptive intervention with trigger detection and a generative chatbot: Smoking cessation use...

Area of Science:

• Digital Health
• Behavioral Science
• Artificial Intelligence

Background:

• Just-in-time adaptive interventions (JITAIs) face challenges in tailoring messages and timing interventions effectively.
• Personalized support is crucial for successful behavior change.

Purpose of the Study:

• To introduce an architecture for JITAIs that integrates message tailoring and timely intervention delivery.
• To leverage large language models (LLMs) for personalized behavior change support.

Main Methods:

• Developed a system combining trigger detection for optimal intervention timing with prompt engineering on an LLM.
• Applied the system as a proof of concept for smoking cessation.
• Evaluated the chatbot's responses through expert interviews for correctness, relevancy, and personalization.

Main Results:

• An expert confirmed the chatbot's support was correct, personal, and tailored to user triggers and circumstances.
• The chatbot's capabilities were deemed effective and acceptable for smoking cessation support.
• Minor suggestions were provided for further enhancement.

Conclusions:

• Large language models (LLMs) with prompt engineering can create personalized, trigger-responsive chatbots for JITAIs.
• Integrating trigger detection and generative chatbots into JITAIs is feasible while maintaining user privacy.