Wearable Meets LLM for Stress Management: A Duoethnographic Study Integrating Wearable-Triggered Stressors and LLM Chatbots for Personalized Interventions

2025-02-24

AI Summary

Current stress management interventions lack timeliness and personalization. Method: This study proposes a closed-loop intervention paradigm integrating wearable physiological sensing with a lightweight fine-tuned large language model (LLM). Real-time heart rate variability (HRV) and electrodermal activity (EDA) signals trigger contextualized intervention events; brief, structured situational descriptions then prompt the LLM to generate personalized coping suggestions. A 22-day dual-ethnographic study evaluated the “trigger–describe–generate” pipeline. Results: Contextualized interventions achieved a 3.2× higher user adoption rate and perceived effectiveness compared to generic advice; 20% of detected physiological anomalies warranted intervention. The work introduces the first deep coupling mechanism between physiological signal processing and semantic understanding, establishing reusable design principles for stress-responsive systems and a multidimensional evaluation framework encompassing physiological, behavioral, and subjective metrics.

Technology Category

• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Humans and AI: Brain-Sensing and Analysis
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

We use a duoethnographic approach to study how wearable-integrated LLM chatbots can assist with personalized stress management, addressing the growing need for immediacy and tailored interventions. Two researchers interacted with custom chatbots over 22 days, responding to wearable-detected physiological prompts, recording stressor phrases, and using them to seek tailored interventions from their LLM-powered chatbots. They recorded their experiences in autoethnographic diaries and analyzed them during weekly discussions, focusing on the relevance, clarity, and impact of chatbot-generated interventions. Results showed that even though most events triggered by the wearable were meaningful, only one in five warranted an intervention. It also showed that interventions tailored with brief event descriptions were more effective than generic ones. By examining the intersection of wearables and LLM, this research contributes to developing more effective, user-centric mental health tools for real-time stress relief and behavior change.

Problem

• Personalized stress management using wearables and LLMs.
• Effectiveness of tailored chatbot interventions for stress.
• Real-time stress relief through wearable-detected physiological prompts.

Innovation

• Wearable-integrated LLM chatbots
• Personalized stress management interventions
• Duoethnographic study for real-time analysis