User Trust in Assisted Decision-Making Using Miniaturized Near-Infrared Spectroscopy

Title of the Paper

Research on User Trust in Decision Support Using Miniature Near-Infrared Spectroscopy Technology

Bibliographic Information

• Subject Area: User Trust and Human-Computer Interaction, Miniature Near-Infrared Spectroscopy (NIRS) Technology, Decision Support
• Keywords: NIRS, Food Detection, Decision Support, User Trust, Visualization Design, Time Analysis, User Experience

Research Background and Issues

• Identified Problems or Challenges:

  1. How to design and implement decision support systems in daily life to assist non-expert users (e.g., detecting allergens such as gluten in food).
  2. Current research primarily focuses on expert scenarios in medical or commercial fields, with limited studies on object recognition tasks in everyday decision-making.
  3. User interface design and the visualization of machine learning models play a critical but underexplored role in decision efficiency and user trust.

• Significance of the Research:

  1. With the global increase in food allergies and health risks, accurately detecting food ingredients is crucial for ensuring consumer safety.
  2. Novel miniature near-infrared spectroscopy (NIRS) technology offers a user-friendly method for non-experts to detect food ingredients, with broad application potential.

• Research Motivation:

  • To explore how to enhance the usability, decision-making efficiency, and user trust in NIRS technology for non-expert users.
  • To investigate potential user behavior biases (e.g., risk aversion) and the impact of decision time and user interface factors.
  • To design decision support systems better suited for everyday scenarios.

Solution

• Proposed Methods or Solutions:

  1. Designed and implemented an experimental protocol using a miniature NIRS scanner to detect whether food items (e.g., tortillas) contain gluten.
  2. Investigated the effects of different packaging label conditions (English, Russian, no label) and three types of information visualization (numerical, progress bar, grid).
  3. Developed a machine learning model to evaluate user decision accuracy, time consumption, and trust behavior based on experimental data.

• Innovations:

  1. Conducted the first analysis of user trust and usage patterns of miniature NIRS technology in everyday decision-making tasks.
  2. Introduced new variables, such as label language and confidence visualization, to study their impact on non-expert users' cognition and trust.
  3. Provided practical recommendations for designing effective interactive interfaces for emerging technologies.

• Implementation Steps and Key Techniques:

  1. Experimental Design:
     • 36 participants tested 18 tortilla samples under different packaging label and visualization conditions.
     • Food samples were scanned using a miniature NIRS device, and the estimated results were displayed.
  2. Machine Learning Model Development:
     • Gluten detection was performed using Random Forest and Support Vector Machine models, with the latter selected for its stability in analysis.
  3. User Behavior Analysis:
     • User inspection time, scanning time, and trust decision behaviors were recorded, and qualitative interviews were conducted to assess factors influencing user trust.

Research Findings

• Specific Results Achieved:

  1. Users' decision accuracy with NIRS technology (0.858) was higher than their independent decision accuracy (0.767), particularly when label information was incomprehensible (0.634) or absent (0.667).
  2. Among the three visualization methods, participants preferred numerical and progress bar formats, while the grid format took longer and was the least favored.
  3. Users were more inclined to trust the NIRS device when label information was limited, especially when their independent decisions conflicted with the scanning results.
  4. Significant risk-averse behavior was observed, with users more likely to report "contains gluten" to avoid potential health risks.

• Advantages Compared to Existing Solutions:

  1. Extended the application of emerging NIRS technology from industrial and scientific laboratory settings to everyday consumer scenarios.
  2. Provided detailed design guidelines for enhancing trust and user experience for non-expert users.
  3. Offered specific time optimization recommendations to improve interaction efficiency.

• Experimental or Evaluation Results:

  • Different label and confidence visualization methods significantly influenced decision time and trust levels.
  • Participants required approximately 75% confidence to trust the device's recommendations.
  • In terms of user inspection time, incomprehensible information (Russian labels) disrupted decision efficiency.

• Limitations and Future Directions:

  1. The experiment was limited to tortillas; future studies should expand to other food detection applications (e.g., peanuts, dairy products).
  2. The current experimental setting was in a laboratory environment; future research should validate results in real-world settings to enhance ecological validity.
  3. The study did not extensively explore the relationship between participants' individual characteristics (e.g., attitudes toward technology) and trust.
  4. Future research could explore additional dimensions of visualization design (e.g., color, hierarchy) to further improve user experience.