Impacts of Personal Characteristics on User Trust in Conversational Recommender Systems

Title of the Paper

Impacts of Personal Characteristics on User Trust in Conversational Recommender Systems

Paper Information

• Subject Areas: Human-Computer Interaction, Conversational Recommender Systems, Trust Modeling
• Keywords: Conversational Recommender Systems, User Trust, Personalization, Personality Traits, Trust Propensity, Mixed-Initiative Interaction, Task Complexity

Research Background and Issues

• Research Background:
  Conversational recommender systems (CRSs) mimic human advisors, assisting users in finding items of interest through multi-turn dialogues. In recent years, CRSs have gained significant attention in domains such as media and e-commerce. CRSs support mixed-initiative interaction, combining user initiative with system initiative, enhancing interaction naturalness and user exploration experience.

• Research Issues and Challenges:

  • Existing research on CRSs overly emphasizes recommendation efficiency and quality, with limited focus on user trust.
  • User trust is a critical factor in determining whether users accept recommendations and continue using the system. Trust levels may vary depending on user-specific characteristics (e.g., personality traits) and system design.
  • There is a lack of research on how initiative strategies in system interaction (user-initiated vs. mixed-initiative) affect user trust.

• Research Motivation and Related Work:

  • Exploring the relationship between user characteristics and CRS design can optimize system design and enhance individual user trust.
  • Based on Hoff and Bashir's three-layer trust model, this study investigates how user-related, system-related, and context-related factors influence user trust.

Solution

• Research Methodology:
  The authors conducted a user study with 148 participants to examine the impact of three types of personal characteristics (personality traits, trust propensity, domain knowledge) on user trust in two types of CRSs (user-initiated and mixed-initiative).

• Innovations:

  1. Systematically explored the interaction effects of user characteristics and initiative strategies on CRS trust for the first time.
  2. Incorporated specific user attributes into trust modeling and proposed design recommendations to optimize CRS interaction methods.
  3. Used structural equation modeling (SEM) and linear regression analysis to quantitatively validate cognitive trust pathways and interaction effects.

• Implementation Steps:

  1. Prototype Design:
     • Developed two recommendation prototypes: a user-initiated system and a mixed-initiative system.
     • The user-initiated system responded to explicit user needs, while the mixed-initiative system combined user feedback with proactive suggestions.
  2. Task Design:
     • Simple task: Find and evaluate suitable songs.
     • Complex task: Explore various music styles and select the top 5 songs from a list of 20.
  3. Participant Recruitment and Measurement:
     • Used questionnaires to assess participants' Big Five personality traits, trust propensity, and music domain knowledge.
  4. Data Analysis:
     • Employed SEM to analyze cognitive pathways of trust formation.
     • Used regression models to explore the interaction effects of user characteristics, system initiative, and task complexity.

Research Findings

• Key Findings:

  1. Direct Effects of User Characteristics on Trust:
     • Users with high conscientiousness showed greater trust and preference for system-initiated suggestions.
     • Extroverted users were more likely to appreciate recommendation quality.
     • Trust propensity and music domain knowledge directly enhanced system trust and usage intention.
  2. Interaction Effects of Initiative Strategies:
     • Mixed-initiative systems were particularly beneficial for users with high conscientiousness.
     • For users with high domain knowledge, user-initiated strategies were more suitable, while mixed-initiative strategies were better for those with lower domain knowledge.
  3. Effects of Task Complexity on Trust:
     • For simple tasks, users focused more on recommendation quality.
     • For complex tasks, trust propensity had a more significant impact.

• Comparison with Existing Solutions:

  • The authors emphasized that "dialogue interaction experience" has a stronger influence on trust than "recommendation quality," addressing the limitations of traditional recommender systems that focus solely on accuracy and explainability.

• Experimental or Evaluation Results:

  • The SEM model validated trust formation pathways, with an improved adjusted R^2 value, demonstrating the model's strong explanatory power for cognitive trust pathways.
  • Regression analysis revealed key interaction effects, such as a significant positive correlation between conscientiousness and mixed-initiative strategies.

• Limitations and Future Directions:

  • The study was limited to the music recommendation domain, and its applicability to other fields (e.g., healthcare, e-commerce) needs further validation.
  • It did not cover research on voice interaction platforms, which could be extended to voice-based CRSs in the future.
  • Other dimensions of trust, such as security and privacy protection, were not comprehensively explored.

Conclusion

This study provides practical, personalized recommendations for CRS design by deeply investigating the effects of user characteristics, initiative interaction strategies, and task complexity on trust. The research significantly advances the field of trust modeling in conversational recommender systems and contributes to the broader domains of human-computer interaction and personalized recommendations.