TasteWeights: A Visual Interactive Hybrid Music Recommender System

BLOG
Submitted by Sharon Hsiao on Wed, 09/12/2012 - 12:08am.
Sharon Hsiao's picture

This paper was just published yesterday in International Conference on Recommender Systems 2012. Normally, traditional recommender systems mimic the “weighting” into sophisticated algorithms. All users get is a rank list of recommended items. This paper talks about the design of using interactive visualization to allow users to adjust the importance of the features, and items which are relevant for the recommendation. The idea is simple and allow users to understand and control aspects of the recommendation process.

TasteWeights is a hybrid recommender. It combines one’s music library and his friends (or other social media) music tastes together to generate recommendations.
Check the following system interface and a demo

This is an explanatory interface which educate users about recommendation systems and enable them to tweak the underlying algorithms in realtime. The study results show that the interface:

  • increases user satisfaction
  • The interactions at recommendation time can improve recommendation accuracy and user experience
  • the hybrid strategies with social outperformed traditional collaborative filtering

    • Do you find the similarities from what we did in today’s EdLab's Development & Research?

    The zoomable researcher recommender interface presented by team Rebecca, Lu, Laura, Hui Soo. Totally spot it ☺

    In the context of music, it may make sense to combine “social element” for the hybrid recommendations. In our context, we can consider more reciprocal recommendations mechanism for bettering matching.

    Posted in Public | Sharon Hsiao's blog | login or register to post comments »



    Janice Joo's picture
    Janice Joo Says:
    Wed, 09/12/2012 - 10:51am

    Hi Sharon,

    Thanks for sharing this really cool article and corresponding TasteWeights recommendation tool! I am impressed that tweaks are updated realtime. It encourages "iterative querying", aka "exploration".

    Could you divulge a little more about how this interface is explanatory?


    login or register to post comments »

    Sharon Hsiao's picture
    Sharon Hsiao Says:
    Wed, 09/12/2012 - 11:26am

    it describes how the system displays the recommendation transparency to the users.

    moreover, the adjustable weights feature really gives the users the control to know and to learn "why" these pieces of music are recommended.

    more like, an initiative and self-explained tool, rather than just a confound engine.


    login or register to post comments »