Electronic Theses and Dissertations

Date

2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Computer Science

Committee Chair

Santosh Kumar

Committee Member

Brandon Booth

Committee Member

David M. Almeida

Committee Member

Myounggyu Won

Abstract

Advances in wearables and AI enable real-time stress detection for JITAI, yet long-term stress reduction remains difficult because interventions must align with the underlying stressors. Traditional methods like annual or end-of-day surveys miss many stressors due to recall bias or participant burden. AI-triggered prompts offer a timely and low-burden alternative for capturing real-world stressors as they occur. To demonstrate the feasibility of using AI-triggered prompts for real-time stressor capture, we conducted the MOODS Study, a 100-day field study with 122 participants wearing smartwatches equipped with a pre-trained stress model. When physiological stress events ended, participants were prompted to log their stress ratings and stressors. Despite no incentives, participants maintained an 81% 30-day retention rate, contributed over 11,000 stressors, and rated the app with 85% approval. High engagement and retention were supported by self-reflective visualizations integrated into a Personal Informatics System. These visualizations helped participants build awareness of their stressors, leading to a 10% reduction in stress intensity, a decrease of 10 stressors per month, and 14 self-initiated behavior changes. To reduce burden further, we introduce AI-Triggered Adaptive Prompting (ATP), which optimizes prompt frequency by accounting for individual variability. Simulation results show ATP balances burden and coverage, enabling efficient data collection. Next, we developed asymptotic models to estimate the latent frequency of different kinds of real-life stressors that address sample sparsity and sampling bias. We find that people experience 5.39 stressors per day, on average. Despite decades of stress research and its importance in HCI, the duration of real-world stressor effects remains largely unknown. We introduce a new approach that models stressor longevity as a time-to-event process using a Weibull survival model. Our findings show that longevity inferred from wearables is significantly shorter than that inferred through self-reports. Finally, survival modeling identified predictors of who is most likely to change behavior, with stress frequency, intensity, and engagement as significant factors. AI-triggered prompts effectively capture real-world stressors in real time without burdening participants, sustaining engagement and even motivating self-initiated behavior change. They also enable new insights into the frequency and duration of stress, advancing stress research and informing more context-aware HCI systems.

Comments

Data is provided by the student.”

Library Comment

Dissertation or thesis originally submitted to ProQuest.

Notes

Open Access

Download

Share

COinS
 

Archival Statement

This item was created or digitized prior to April 24, 2026, or is a reproduction of legacy media created before that date. It is preserved in its original, unmodified state specifically for research, reference, or historical recordkeeping. This material is part of a digital archival collection and is not utilized for current University instruction, programs, or active public communication. In accordance with the ADA Title II Final Rule, the University Libraries provides accessible versions of archival materials upon request. To request an accommodation for this item, please submit an accessibility request form.