Interactive music genre exploration with visualization and mood control

Title of the Paper

Interactive Music Genre Exploration with Visualization and Mood Control

Paper Information

• Domain: Music recommendation systems and user interface design
• Keywords: Recommendation systems, interaction design, visualization, exploration, music, mood, user study

Research Background and Problem

• Research Problem and Challenges: Recommendation systems help users discover new items and explore new preferences, but there is limited research on how to improve users' understanding and acceptance of recommended content, as well as how to support users in exploring unfamiliar domains. When exploring unfamiliar music genres, users may find it difficult to understand the relationship between the recommended content and their personal preferences, which reduces their interest in exploration.

• Significance: In the context of mitigating the "filter bubble" effect, supporting users in exploring and accepting new preferences in recommendation systems becomes particularly important. If users cannot associate recommended content with their existing preferences, it discourages them from further exploration.

• Research Motivation and Related Work: Visualization and user control are important methods for improving the transparency and interpretability of recommendation systems. Interactive visualization of the relationship between users and new music genres can help users understand the recommendation process. Additionally, mood is a key factor influencing music preferences. Incorporating mood slider controls can further enhance users' perception of and interest in exploring recommended content.

Solution

• Methods and Solutions: This study designs and compares two different visualization methods:

  1. Bar Chart Visualization: Suitable for explaining recommended content, allowing users to directly compare recommended tracks with their current preferences.
  2. Contour Plot Visualization: Displays recommended content, new music genres, and user preferences within a two-dimensional mood space (energy and valence), offering more comprehensive information.

  Additionally, a mood control slider is introduced, enabling users to adjust the emotional characteristics (energy and valence) of the recommended tracks.

• Innovations:

  1. Combining two visualization techniques with mood control to provide users with a highly transparent and operable interactive experience.
  2. The contour plot offers an integrated view, showing the distribution relationships between recommendations, user preferences, and new music genres.
  3. The inclusion of a mood control slider allows users to adjust recommendations from an emotional perspective, further enhancing personalization.

• Implementation Steps and Techniques:

  1. Extract users' Spotify listening data to build a Gaussian Mixture Model (GMM)-based user preference model.
  2. Provide three recommendation algorithms (personalized, non-personalized, hybrid), ultimately adopting the hybrid method to balance personalized and new music genre needs.
  3. In the visualization interface, users adjust mood preferences via sliders while observing changes in recommended content through bar charts or contour plots.

Research Findings

• Specific Findings:

  • Users found the contour plot more informative than the bar chart, helping them better understand recommended content and new music genres.
  • The contour plot with mood control performed best in aiding users to explore new music genres.
  • Users spent more time on the contour plot and used the mood slider more frequently.
  • Regardless of the visualization method, participants reported improved understanding and sense of control over the recommended content.

• Advantages Compared to Existing Solutions:

  • The contour plot enhances the transparency of recommendations while helping users understand how the system balances their preferences with target music genres.
  • The combination of the mood slider and contour plot further increases interactivity and the depth of user exploration.

• Experimental or Evaluation Results:

  • A user study involving 102 participants tested four conditions: without mood slider and with slider, bar chart and contour plot. Analysis showed:
    • The contour plot significantly improved users' perceived control, information richness, and comprehensibility of the system, thereby enhancing the overall exploration experience.
    • The effect of mood control was more pronounced in the contour plot condition.
    • Users' interaction behaviors (e.g., slider usage frequency and exploration time) supported these subjective improvements.

• Limitations and Future Directions:

  • Limitations: Participants were primarily university students, whose ability to understand complex visual data may be stronger than that of the general population.
  • Future Directions: Extend the study to diverse user groups to validate the effectiveness of the approach across a broader audience. Additionally, further research is needed to simplify the presentation of contour plot information, making it suitable for users with lower educational backgrounds.