Title of the Paper

DeepStress: Supporting Stressful Context Sensemaking in Personal Informatics Systems Using a Quasi-experimental Approach

Paper Information

• Subject Area: Personal informatics systems and mental health, aiding users in managing daily stress through causal analysis.
• Keywords: Personal informatics systems, mental health, quasi-experimental methods, causality, data-driven reflection.

Research Background and Problem

• Identified Problems or Challenges:

  • While personal informatics (PI) systems guide users in deriving insights from self-monitoring data, they remain limited in helping users understand causal relationships.
  • The complexity of data and biases from external confounding factors may lead to erroneous conclusions when analyses are solely based on correlations.
  • Although self-experimentation methods are rigorous, they are challenging to implement in daily life.
  • There is a lack of PI systems that apply quasi-experimental methods to assist in causal analysis.

• Importance of the Problem:

  • Clearer causal analysis can help users take effective, long-term stress management actions based on data, thereby improving mental health.

• Research Motivation and Related Work:

  • Literature and preliminary interviews reveal that users need more intuitive and accurate identification of contexts affecting stress.
  • Quasi-experimental methods are considered a practical and scientifically valid solution for controlling external confounding factors.

Solution

• Proposed Method or Solution:

  • Design the DeepStress system, which helps analyze context-related causal relationships with stress based on six weeks of users' self-collected data.
  • Employ matching techniques from quasi-experimental methods to balance confounding variables between treatment and control groups for causal analysis.

• Innovations:

  • Utilize quasi-experimental methods to infer causal relationships from observational user data without the need for strict randomized experiments.
  • Introduce a two-stage causal analysis to not only determine whether a context is related to stress but also explore other factors within the given context that might influence stress.

• Implementation Steps and Key Techniques:

  1. Data Collection: Participants use a mobile app to record stress, activities, locations, social environments, etc., through the experience sampling method.
  2. Causal Analysis Method: Perform context matching using Coarsened Exact Matching (CEM) to balance confounding variables.
  3. System Features:
     • Historical Stress Data Navigation: Users can view calendars and charts to revisit past records.
     • Causal Relationship Results Display: Visualizations and explanations directly inform users about stress-related causal contexts.
     • Support for Planning and Action: Demonstrates how to regulate stress within specific contexts.

Research Outcomes

• Specific Outcomes:

  • DeepStress was proven to help users review historical data, identify stress-related contexts, and understand causal relationships.
  • Users improved their behavior planning and took targeted actions for stress management through the system.
  • The study revealed users' cognitive processes and challenges in dealing with causal relationships.

• Advantages Over Existing Solutions:

  • Compared to systems relying solely on correlation analysis, DeepStress is more scientifically robust in causal inference.
  • The system enhances users' understanding of complex relationships between contexts, rather than focusing on isolated single contexts.

• Experimental or Evaluation Results:

  • In-depth user studies showed a user experience score of 74.1 for DeepStress, indicating good usability in supporting causal relationship exploration.
  • Users independently generated interpretations through the system, enhancing their understanding of results and better planning their daily lives.

• Limitations and Future Directions:

  • Challenges in recording fine-grained contexts: Providing more detailed options or supporting user annotations may improve accuracy.
  • Data demands and burden: Automating data collection or customizing data collection scopes could reduce user burden.
  • Limitations of full-sample matching: Exploring other quasi-experimental methods or extending to groups with diverse backgrounds.
  • Future research directions include comparing systems based on causal analysis with other PI systems to evaluate their relative advantages.

Acknowledgments

• This research was supported by the Korean government (MSIT) as part of the development project for biological and medical technologies.