Experiencer, Helper, or Observer: Online Fraud Intervention for Older Adults Through a Role-based Simulation Approach

Paper Title

Experiencer, Helper, or Observer: Online Fraud Intervention for Older Adults Through a Role-based Simulation Approach

Publication Info

• Topic area: Online fraud education for older adults using role-based simulation.
• Keywords: Online fraud, older adults, role-based simulation, fraud cues, LLM-based intervention, Experiential Learning Theory, Learning by Teaching, Social Learning Theory, fraud awareness, anti-fraud education.

Background and Problem

• Problem / challenge: Existing anti-fraud education for older adults relies on passive, traditional formats that lack interactivity and fail to address the complexity of real-world fraud scenarios. Most interventions focus solely on victims, neglecting other roles such as helpers and observers.
• Significance: Older adults are disproportionately targeted by online fraud, with significant financial losses and emotional distress. Effective fraud education is critical to empower them to recognize and resist scams.
• Motivation and related work: Prior studies have explored role-based simulations for phishing and cybersecurity education but have not tailored these approaches to older adults or examined multi-role perspectives. This paper addresses the gap by introducing a role-based simulation framework specific to online fraud education for older adults.

Solution

• Proposed approach: ROLESafe, a role-based simulation intervention where older adults learn fraud awareness through three roles: Experiencer, Helper, and Observer.
• Novelty:
  1. Development of a tailored online fraud intervention using role-based simulation for older adults.
  2. Introduction of multi-role perspectives (Experiencer, Helper, Observer) to enhance learning outcomes.
  3. Empirical evaluation of role-based simulation effectiveness in fraud cue identification and user experience.
• Procedure and key techniques:
  • Design of three learning modes:
    • Experiencer: Participants engage directly with a simulated scammer (GPT-4o).
    • Helper: Participants assist a simulated victim in identifying fraud cues.
    • Observer: Participants passively review scam conversations generated by GPT-4o.
  • Fraud cues were extracted from real-world cases targeting older adults in China, focusing on financial and health vulnerabilities.
  • A between-subjects experiment with 144 older adults evaluated the intervention’s effectiveness using fraud cue identification tests, self-efficacy scales, and user experience surveys.

Results

• Concrete findings:
  • Fraud cue identification (F1 scores): Experiencer (0.81), Helper (0.85), Observer (0.76), Control (0.66).
  • Self-efficacy scores: Experiencer (4.41), Helper (4.40), Observer (4.26), Control (4.04).
  • Cue coverage rates during learning: Helper (90% for all cues), Experiencer (34.85% for all cues), Observer (100% for all cues).
• Advantage over baselines:
  • Role-based simulation groups significantly outperformed the control group in fraud cue identification (P < 0.001).
  • Helper role demonstrated the highest effectiveness, outperforming Observer (P = 0.025).
• Experiments / evaluation:
  • Participants: 144 older adults in China (aged 50+).
  • Metrics: Fraud cue identification (F1 scores), self-efficacy, user experience (enjoyment, usability, future behavioral intention).
  • Follow-up study with 84 participants showed stable scores but no significant group differences.
• Limitations and future work:
  • Cultural specificity: Findings may not generalize beyond the Chinese context.
  • Long-term effectiveness: Follow-up results did not replicate initial group differences.
  • Role-specific challenges: Experiencer role caused worry; Helper role led to emotional burden when persuasion failed.
  • Future directions: Incorporating multimodal fraud cues, dynamic updates via AI, and longitudinal evaluations.

Summary

This study introduces ROLESafe, a role-based simulation framework for online fraud education tailored to older adults. By engaging participants as Experiencers, Helpers, or Observers, the intervention significantly improved fraud cue identification compared to traditional methods. Helpers demonstrated the highest effectiveness, highlighting the value of active roles in learning. While LLM-based simulations proved feasible and well-received, challenges such as emotional burden and cultural specificity remain. Future work should explore scalable, multimodal, and longitudinal approaches to enhance fraud education across diverse populations.