Perception of Fairness in Group Music Recommender Systems

Title of the Paper

Perception of Fairness in Group Music Recommender Systems

Paper Information

• Domain: Collaborative recommender systems, specifically fairness research in group music recommendation
• Keywords: Group recommender systems, fairness, personalization, user study, music recommender systems, Spotify API, Big Five personality traits, ranking algorithms, user perception, experimental design

Research Background and Problem

• Research Problem: Group recommender systems (GRS) need to balance the diverse preferences of multiple users to ensure fairness in recommendations. However, how individual user personalities influence their perception of fairness remains underexplored.
• Research Importance: In group recommendations, users' perception of system fairness directly affects satisfaction and user experience. Therefore, understanding the impact of personality traits on fairness perception is crucial.
• Research Motivation: Existing studies focus on algorithms and optimization techniques but fail to address the relationship between personality traits and fairness perception. This study uses music recommendation as the application scenario to fill this research gap.

Solution

• Proposed Solution:

  1. Developed a group music recommender system based on the Spotify API.
  2. Designed two ranking algorithms: a time-based ranking algorithm and a dissimilarity-based ranking algorithm.
  3. Conducted an experiment using both ranking algorithms to investigate users' perception of fairness.

• Innovations:

  1. First study to explore the impact of user personality traits on fairness perception in group recommender systems.
  2. Collected direct feedback through user experiments, providing empirical data for recommender system design.

• Implementation Steps and Techniques:

  • Utilized the Spotify Web API to provide personalized and group recommendations.
  • Calculated song rankings based on user preferences, including:
    • Time-based algorithm (sorted by addition time).
    • Content-based dissimilarity algorithm (using Spotify track attributes to compute Euclidean distance and sorting by dissimilarity scores).
  • Integrated a voting mechanism to enhance user engagement.
  • Administered the Big Five personality questionnaire to assess users' personality traits.
  • Used surveys to evaluate users' perception of fairness and its relationship with the algorithms.

Research Findings

• Specific Findings:

  1. Openness personality trait negatively correlates with the importance of fairness in group recommendations.
  2. The time-based ranking algorithm (TIM) was perceived as less fair by users, while the dissimilarity-based ranking algorithm (DIS) performed better.
  3. Significant relationships between personality traits and fairness perception were confirmed.

• Advantages and Comparisons:

  • The dissimilarity-based algorithm was more favorably perceived in terms of fairness compared to the time-based algorithm.
  • Users with high openness tend to prioritize discovering new music over fairness.

• Experimental Results:

  1. The average minimum group proportion for the DIS algorithm was higher than that of the TIM algorithm, further demonstrating its superior fairness performance.
  2. Users with openness and extroversion traits were more inclined to use the recommended playlists outside the group context.

• Limitations and Future Directions:

  1. The study only investigated two ranking algorithms; future research should expand to more types and complexities of ranking algorithms.
  2. The impact of long-term group interactions and relationships on fairness perception remains underexplored and could be a focus for future work.

Conclusion

This study reveals the significant relationship between personality traits and fairness in group recommender systems and validates the advantages of certain ranking algorithms in enhancing users' fairness perception. Future research could further explore how group dynamics influence recommender system design and how intelligent algorithms can optimize fairness and user experience in recommendations.