Fighting Choice Overload: Using Recommender Systems for a Better Experience - EuroIA2012, Rome

Transcript

1. Fighting Choice Overload
   Using Recommender Systems
   for a Better Experience
   Pónya Judit
   FatDUX Budapest
   EuroIA 2012, Rome
   September 29, 2012.
   

   View Slide

2. Judit Pónya
   @ptxster
   

   View Slide

3. find
   search
   

   View Slide

4. ! !
   

   View Slide

5. Aspects
   ■ Recommendation
   instead of “word-of-mouth”
   

   View Slide

6. Aspects
   ■ Recommendation
   instead of “word-of-mouth”
   ■ Personalization
   as users have different preferences and goals
   

   View Slide

7. Aspects
   ■ Recommendation
   instead of “word-of-mouth”
   ■ Personalization
   as users have different preferences and goals
   ■ Content discovery
   Help users to find good or useful things without
   specifically searching for them
   

   View Slide

8. Aspects
   ■ Recommendation
   instead of “word-of-mouth”
   ■ Personalization
   as users have different preferences and goals
   ■ Content discovery
   Help users to find good or useful things without
   specifically searching for them
   ■ Fighting choice overload
   Choosing from many alternatives is not an easy task...
   Don’t let your users give up!
   

   View Slide

9. When Choice is Demotivating: Can One Desire Too Much of a Good Thing? Iyengar, S. S., & Lepper, M.
   

   View Slide

10. 3%
    30%
    When Choice is Demotivating: Can One Desire Too Much of a Good Thing? Iyengar, S. S., & Lepper, M.
    

    View Slide

11. on shelf!
    tasting
    booth
    

    View Slide

12. View Slide

13. Don’t let them give up!
    ■make it less work to find things
    ■provide some solution
    as soon as possible
    ■make it easier to choose
    

    View Slide

14. 1.
    The human touch
    

    View Slide

15. Business Control
    & Analytics
    Knowledge Base
    User Interface
    Knowledge
    Processing
    Application
    

    View Slide

16. User Interface
    Business
    Control &
    Analytics
    Knowledge Base
    Processing
    Francisco J. Martin (Strands), RecSys2009
    

    View Slide

17. 2. Focus
    on content
    

    View Slide

18. Characteristics of an item
    Build profile
    for each item/user
    based on their
    properties
    item
    

    View Slide

19. Change the focus
    item
    !
    

    View Slide

20. User Behavior
    Looking for clues
    to understand
    preferences & intent
    

    View Slide

21. 3. Focus
    on interaction
    

    View Slide

22. “YouTube Comes To A 5-Star Realization:
    Its Ratings Are Useless”
    1 2 3 4 5
    http://techcrunch.com/2009/09/22/youtube-comes-to-a-5-star-realization-its-ratings-are-useless/
    

    View Slide

23. Events
    ■ 1-5 star rating
    each user would rate
    only a very small portion
    of all items
    

    View Slide

24. Events
    ■ 1-5 star rating
    implicit
    feedback
    each user would rate
    only a very small portion
    of all items
    ■ View page.
    ■ Read article for 1-5 minutes.
    ■ Come back to page.
    ■ Put to cart.
    ■ Buy.
    

    View Slide

25. Don’t make me work.
    

    View Slide

26. Cold start
    problem
    oh, wait....
    

    View Slide

27. Indi Young http://www.flickr.com/photos/indiyoung/7889928594
    I spent like 40 minutes rating movies for Netflix
    a while ago. When I asked for suggestions, I got
    this message. Wow, it made my heart drop.
    I invested all that time, and my preferences were
    not something Netflix seemed to care about.
    

    View Slide

28. " Data
    # Data Processing
    $
    Control
    % UI
    

    View Slide

29. Knowledge Base
    Data quality and
    quantity matters.
    

    View Slide

30. Knowledge Processing Application
    several algorithms in use
    ■build model based on data
    ■make predictions using
    the model
    

    View Slide

31. Business Control
    

    View Slide

32. User Interface
    

    View Slide

33. Business Control & Analytics
    

    View Slide

34. How will itt appear
    on the user interface?
    ■placement
    ■visual design
    ■good title, giving some explanation
    

    View Slide

35. Number of recommendations
    

    View Slide

36. Number of recommendations
    choosability
    Stefano Bussolon, EuroIA2011
    

    View Slide

37. What’s the dealbreaker?
    item
    #3278
    item
    #6723
    item
    #5195
    item
    #4981
    

    View Slide

38. Don’t let it slow
    down your system!
    system performance
    response time (milliseconds)
    Will it handle millions of
    recommendation request?
    Do you have a fallback solution?
    

    View Slide

39. Widgets & other uses
    newsletters
    ■personalized content
    ■target special user group
    site search
    ■modify ranking
    ?
    

    View Slide

40. User Interface
    Business
    Control &
    Analytics
    Knowledge Base
    Processing
    Francisco J. Martin (Strands), RecSys2009
    

    View Slide

41. Don’t let them give up!
    ■make it less work to find things
    ■provide some solution
    as soon as possible
    ■make it easier to choose
    

    View Slide

42. Judit Pónya
    @ptxster
    jponya@fatdux.com
    www.fatdux.com
    Köszönöm!
    

    View Slide

43. View Slide