Work in progress#health
StressTalk: Designing and Evaluating a Conversational Framework Integrating Passive Sensing for Stress Self-Reporting
TL;DRSmartwatches can flag moments that look stressful, but only the person knows what actually happened and how it felt. StressTalk is an LLM-based conversational agent that uses passive sensing as recall cues, so people can describe their daily stress in their own words.
Abstract
Wearable devices can detect potential stress episodes from physiological signals, but these signals are difficult to interpret without the person's own description of what happened and how they felt. We present StressTalk, a framework in which an LLM-based conversational agent conducts a prospective check-in each morning and a retrospective check-in each evening, using potential stress episodes and context inferred from passive sensing data by a multi-agent sensemaking pipeline as recall cues. We are evaluating StressTalk in a multi-week field deployment comparing it with ecological momentary assessment (EMA), and with domain experts who interpret the resulting self-reports. This work is in progress; more details will follow.
