Auralist: Introducing Serendipity into Music Recommendation  @danielequercia
U   C   L <who am i>
U   C   L daniele quercia
U   C   L
U   C   L
U   C   L
U   C   L
 
 
o ffline & online
Introducing serendipity in recommendations
Introducing serendipity in recommendations
 
 
 
F ilter bubble  (chilling idea  … for some) Your content limited by your past& self-propagating interests
<ul><li>Goal:  how to produce recommendations that are </li></ul><ul><li>Accurate </li></ul><ul><li>Diverse </li></ul><ul>...
<ul><li>Auralist:  framework broadening musical horizons ;) </li></ul><ul><li>Basic </li></ul><ul><li>Community-Aware </li...
<ul><li>Basic </li></ul><ul><li>Employs Latent Dirichlet Allocation (LDA) </li></ul>
LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur mo...
LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur mo...
LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur mo...
LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur mo...
LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur mo...
LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur mo...
LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur mo...
1. Basic Auralist match( user’s history, artist )
2. Community-Aware balance  *  match( user’s history, artist ) *  diversity( artist )
2. Community-Aware balance  *  match( user’s history, artist ) *  diversity( artist ) favors  artists with  broader fan ba...
2. Community-Aware balance  *  match( user’s history, artist ) *  diversity( artist ) favors  artists with  broader fan ba...
3. Bubble-Aware The Beatles HolyBlood The Rolling Stones balance  *  match( user’s history, artist ) *  bubbleness( artist )
3. Bubble-Aware balance  *  match( user’s history, artist ) *  bubbleness( artist ) favors  cluster-avoiding  artists by p...
4. Full Auralist Rank interpolation of 1. 2. and 3.
d o they work?
 
+ + - - -
+ - - + +
Both  improve  novelty, diversity and serendipity  b ut with accuracy loss  OK news!
Good news:  accuracy loss can be minimised good bad
Good news:  accuracy loss can be minimised good bad
User Study:  Basic Auralist vs. Full Auralist Serendipity Enjoyment
User Study:  Basic Auralist vs. Full Auralist Some:  accept accuracy loss for serendipity Majority:  favours of greater ac...
So what?
Future  (well, current & you could help)
1.  Nudging Now:  Auralist Next:  ‘Nudge’ people for serendipity
social media  language personality social media 2.  Personality
language personality social media 2.  Personality @ CSCW
3.  Why’s
2  personality 1   nudging 2  why’s
2  personality 1   nudging 2  why’s
@danielequercia
Upcoming SlideShare
Loading in …5
×

Auralist: Introducing Serendipity into Music Recommendation

1,285 views
1,175 views

Published on

Recommendation systems exist to help users discover content in a large body of items. An ideal recommendation system should mimic the actions of a trusted friend or expert, producing a personalised collection of recommendations that balance between the desired goals of accuracy, diversity, novelty and serendipity. We introduce the Auralist recommendation framework, a system that - in contrast to previous work - attempts to balance and improve all four factors simultaneously. Using a collection of novel algorithms inspired by principles of ‘serendipitous discovery’, we demonstrate a method of successfully injecting serendipity, novelty and diversity into recommendations whilst limiting the impact on accuracy. We evaluate Auralist quantitatively over a broad set of metrics and, with a user study on music recommendation, show that Auralist’s emphasis on serendipity indeed improves user satisfaction.

Published in: Technology, News & Politics
0 Comments
3 Likes
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
1,285
On SlideShare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
26
Comments
0
Likes
3
Embeds 0
No embeds

No notes for slide
  • notes
  • notes
  • notes
  • notes
  • notes
  • notes
  • Auralist: Introducing Serendipity into Music Recommendation

    1. 1. Auralist: Introducing Serendipity into Music Recommendation @danielequercia
    2. 2. U C L <who am i>
    3. 3. U C L daniele quercia
    4. 4. U C L
    5. 5. U C L
    6. 6. U C L
    7. 7. U C L
    8. 10. o ffline & online
    9. 11. Introducing serendipity in recommendations
    10. 12. Introducing serendipity in recommendations
    11. 16. F ilter bubble (chilling idea … for some) Your content limited by your past& self-propagating interests
    12. 17. <ul><li>Goal: how to produce recommendations that are </li></ul><ul><li>Accurate </li></ul><ul><li>Diverse </li></ul><ul><li>Novel </li></ul><ul><li>Serendipitous </li></ul>
    13. 18. <ul><li>Auralist: framework broadening musical horizons ;) </li></ul><ul><li>Basic </li></ul><ul><li>Community-Aware </li></ul><ul><li>Bubble-Aware </li></ul><ul><li>Full </li></ul>
    14. 19. <ul><li>Basic </li></ul><ul><li>Employs Latent Dirichlet Allocation (LDA) </li></ul>
    15. 20. LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur more than chance: keep them in the bin else: put them into another bin (@ random) Facebook Twitter
    16. 21. LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur more than chance: keep them in the bin else: put them into another bin (@ random) Facebook Twitter social econometrics
    17. 22. LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur more than chance: keep them in the bin else: put them into another bin (@ random) Facebook Twitter social econometrics
    18. 23. LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur more than chance: keep them in the bin else: put them into another bin (@ random) For each doc:
    19. 24. LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur more than chance: keep them in the bin else: put them into another bin (@ random) For each doc: users user1, user2, … (who belong to a given community) artist
    20. 25. LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur more than chance: keep them in the bin else: put them into another bin (@ random) The Beatles HolyBlood
    21. 26. LDA create virtual bins (latent topics) assign words to a bin (@ random) for each bin: select pair of words if co-occur more than chance: keep them in the bin else: put them into another bin (@ random) The Beatles HolyBlood diversity() diversity() similarity()
    22. 27. 1. Basic Auralist match( user’s history, artist )
    23. 28. 2. Community-Aware balance * match( user’s history, artist ) * diversity( artist )
    24. 29. 2. Community-Aware balance * match( user’s history, artist ) * diversity( artist ) favors artists with broader fan bases e.g., The Beatles over HolyBlood
    25. 30. 2. Community-Aware balance * match( user’s history, artist ) * diversity( artist ) favors artists with broader fan bases e.g., The Beatles over HolyBlood … but discounting for popularity
    26. 31. 3. Bubble-Aware The Beatles HolyBlood The Rolling Stones balance * match( user’s history, artist ) * bubbleness( artist )
    27. 32. 3. Bubble-Aware balance * match( user’s history, artist ) * bubbleness( artist ) favors cluster-avoiding artists by pushing the boundaries of a user’s taste
    28. 33. 4. Full Auralist Rank interpolation of 1. 2. and 3.
    29. 34. d o they work?
    30. 36. + + - - -
    31. 37. + - - + +
    32. 38. Both improve novelty, diversity and serendipity b ut with accuracy loss OK news!
    33. 39. Good news: accuracy loss can be minimised good bad
    34. 40. Good news: accuracy loss can be minimised good bad
    35. 41. User Study: Basic Auralist vs. Full Auralist Serendipity Enjoyment
    36. 42. User Study: Basic Auralist vs. Full Auralist Some: accept accuracy loss for serendipity Majority: favours of greater accuracy * serendipity IS a user-specific parameter
    37. 43. So what?
    38. 44. Future (well, current & you could help)
    39. 45. 1. Nudging Now: Auralist Next: ‘Nudge’ people for serendipity
    40. 46. social media language personality social media 2. Personality
    41. 47. language personality social media 2. Personality @ CSCW
    42. 48. 3. Why’s
    43. 49. 2 personality 1 nudging 2 why’s
    44. 50. 2 personality 1 nudging 2 why’s
    45. 51. @danielequercia

    ×