Clarifying or Complicating?: Understanding Older Adults' Engagement with Real-World XAI in E-Commerce

Paper Title

Clarifying or Complicating?: Understanding Older Adults' Engagement with Real-World XAI in E-Commerce

Publication Info

• Topic area: Explainable AI (XAI) in e-commerce, focusing on older adults.
• Keywords: Explainable AI, recommender systems, older adults, e-commerce, personalization, trust calibration, transparency, user-model dashboard, global explanations, local explanations.

Background and Problem

• Problem / challenge: Most XAI research has focused on younger, tech-savvy users, leaving open questions about how older adults engage with explainability features in real-world settings. Older adults often overlook or misinterpret these features, which limits their effectiveness.
• Significance: Understanding older adults’ interaction with XAI is crucial as their participation in digital marketplaces grows. Effective XAI can improve trust, transparency, and user satisfaction in e-commerce systems for this demographic.
• Motivation and related work: Prior studies have primarily tested XAI prototypes in controlled environments, focusing on Western, AI-literate participants. This leaves gaps in understanding how diverse populations, including older adults, engage with deployed XAI systems. The paper builds on research about trust calibration, personalization, and older adults’ attitudes toward AI.

Solution

• Proposed approach: A qualitative study of older adults’ engagement with three types of XAI features—global explanations, local explanations, and a user-model dashboard—on NAVER Shopping, a major e-commerce platform in South Korea.
• Novelty:
  1. Empirical insights into older adults’ experiences with explainable recommender systems in a real-world context.
  2. Identification of three recurring tensions in explanation design for older adults: awareness, evaluation, and transparency.
  3. Design strategies to address these tensions and create more inclusive XAI systems.
• Procedure and key techniques:
  • Conducted semi-structured interviews and think-aloud sessions with 20 older adults aged 60+.
  • Participants explored XAI features (global explanations, local explanations, and user-model dashboard) on their own smartphones.
  • Reflexive thematic analysis was used to identify patterns in participants’ perceptions and interactions.

Results

• Concrete findings:
  • Personalized recommendations were often unnoticed or misinterpreted as advertisements.
  • Global explanations increased awareness but polarized trust—some participants deferred to algorithmic authority, while others dismissed them as marketing rhetoric.
  • Local explanations grounded in user behavior improved relevance assessments and recalibrated skepticism.
  • User-model dashboards enhanced understanding and control but also triggered concerns about surveillance and data exposure.
• Advantage over baselines: The study highlights how layered XAI features (global, local, and profile-level explanations) can address diverse user needs, offering insights into personalization, trust calibration, and transparency.
• Experiments / evaluation:
  • Participants were recruited from South Korea’s Daangn Market and NAVER BAND platforms.
  • Study design included baseline interviews and guided interactions with XAI features.
  • Data analysis involved 1,428 minutes of audio recordings and thematic coding.
• Limitations and future work:
  • Limited generalizability due to the small sample size and cultural context (South Korea).
  • Focused on a single platform, NAVER Shopping; future studies should compare across platforms.
  • Did not examine longitudinal changes in user perceptions; future work should explore sustained engagement.
  • Proposed design strategies are interpretive and require experimental validation.

Summary

This study investigates older adults’ engagement with layered XAI features in NAVER Shopping, revealing diverse mental models and mixed responses to personalization, trust, and transparency. It identifies three key tensions—awareness, evaluation, and transparency—that shape older adults’ experiences with XAI. The findings inform design strategies to make personalization cues clearer, connect system-level explanations to item-level rationales, and present transparency as a controllable resource. These insights contribute to the development of inclusive and adaptive XAI systems that align with older adults’ reasoning processes, goals, and comfort boundaries.