CounterStress: Enhancing Stress Coping Planning through Counterfactual Explanations in Personal Informatics

Research Background and Issues

• Identified Problems and Challenges:

  • While stress management systems help users understand their stress levels, they lack effective and actionable coping strategies in practical applications.
  • The causes of stress often involve multiple factors, and some stress sources (e.g., study or work) cannot be easily avoided.
  • Existing interventions, such as meditation or stretching, are helpful but lack sufficient contextual relevance and personalization.

• Significance:

  • Stress management is crucial for mental health and overall quality of life.
  • Providing users with feasible strategies that can be implemented in real-life situations enhances self-reflection and stress management capabilities.

• Research Motivation and Related Work:

  • Previous Personal Informatics (PI) systems primarily focused on data collection and stress detection, with limited exploration of how to generate specific, context-relevant coping strategies for users.
  • Drawing inspiration from explainability methods in machine learning, particularly counterfactual explanations, may offer more targeted guidance for stress-related scenarios.

Solution

• Proposed Method or Solution:

  • CounterStress is a newly designed PI system that generates personalized stress coping strategies through counterfactual explanations.
  • The system integrates users' historical data and simulates "changes in certain contextual factors" to derive stress-relief strategies.

• Innovations:

  • Introducing counterfactual explanations to answer the question: "How can the situation be adjusted to achieve the desired stress level?"
  • The system provides users with multiple coping strategies, each detailing the contextual factors that need to be modified, allowing users to select the strategy most suitable for their circumstances.
  • Employing machine learning models to establish the relationship between context and stress, thereby offering more personalized recommendations.

• Implementation Steps and Key Technologies:

  1. Data Collection and Processing:
     • Collect users' stress ratings in different contexts using the Experience Sampling Method (ESM).
     • Categorize stress levels into "high" and "low" to simplify analysis.
  2. Counterfactual Generation:
     • Use machine learning models (e.g., Random Forest) to estimate whether a specific context is classified as "high stress."
     • Apply optimization algorithms to identify the minimal feature changes required to generate counterfactual scenarios.
  3. Coping Strategy Recommendation:
     • Based on users' stress patterns, the system generates multiple strategies, such as changing time, location, or social environment.
     • Provide tools for users to compare and select strategies, such as sorting and filtering functionalities.
  4. User Interface Design:
     • The system is divided into three screens: Review, Analysis, and What-If, supporting users in gradually exploring stress sources and coping solutions.

Research Outcomes

• Specific Outcomes:

  • The CounterStress system enables users to understand how to reduce stress by adjusting contexts without requiring complex data analysis.
  • It offers multiple counterfactual strategies for users to choose from and uses quantitative metrics (e.g., historical frequency, stress probability, number of contextual changes) to help users evaluate coping plans.

• Comparison with Existing Solutions and Advantages:

  • Compared to traditional PI systems, CounterStress provides actionable and personalized strategies, addressing the limitation of traditional systems that "only display data without offering actionable advice."
  • The system generates strategies based on users' historical data, enhancing the relevance and specificity of the results.
  • It offers diverse strategies, allowing users to flexibly choose solutions that best suit their needs.

• Experimental or Evaluation Results:

  • Experimental Design:
    • Experiment 1: Laboratory user study to evaluate system usability and user perception.
    • Experiment 2: Field study where users applied the system in real-life scenarios and maintained diaries.
  • Specific Results:
    • The system's usability scored an average of 75 (SUS scale), reflecting a "good" level.
    • Participants reported being able to efficiently plan coping strategies through the system without requiring complex data analysis.
    • In real-world settings, multiple users provided feedback indicating the system positively influenced their stress management behaviors and decision-making.

• Limitations and Future Directions:

  1. The sample size was small (12 participants), requiring larger-scale experiments to further validate the system's effectiveness.
  2. Data variables primarily used categorical data; future research could incorporate continuous variables (e.g., step count or sleep duration) to generate more refined counterfactual explanations.
  3. The system's application in other health domains (e.g., blood sugar management) could be explored, along with enhancing user feedback and integrating group data.

By introducing counterfactual explanations and deep user engagement, CounterStress provides an innovative solution to the complex issue of stress management while demonstrating its practical value in real-world scenarios.