Feature as a Service at Data Labs

Chaerim Yeo
LINE Machine Learning Team Senior Software Engineer
https://linedevday.linecorp.com/jp/2019/sessions/C1-5

Transcript

1. 2019 DevDay
   Feature as a Service at Data Labs
   > Chaerim Yeo
   > LINE Machine Learning Team Senior Software Engineer
   

   View Slide

2. LINE PLATFORM
   

   View Slide

3. DATA LABS
   Sticker
   Data Labs
   Ad
   Manga
   Music
   Live
   News
   > Independent from service/dev depts.
   > Aggregate data across various services
   > Provide analysis/solution from data
   across various services
   

   View Slide

4. Feature as a Service
   

   View Slide

5. WHAT IS IT?
   

   View Slide

6. WHAT IS IT?
   

   View Slide

7. WHAT IS IT?
   

   View Slide

8. WHAT IS IT?
   Standardization Democratization
   

   View Slide

9. AVAILABLE FEATURES
   Z-Features
   User

   Features
   Y-Features
   Obfuscated 

   User Features
   C-Features
   Content
   Features
   

   View Slide

10. Background
    

    View Slide

11. SYSTEM OVERVIEW
    

    View Slide

12. SYSTEM OVERVIEW
    

    View Slide

13. SYSTEM OVERVIEW
    

    View Slide

14. SYSTEM OVERVIEW
    

    View Slide

15. SYSTEM OVERVIEW
    

    View Slide

16. SYSTEM OVERVIEW
    

    View Slide

17. SYSTEM OVERVIEW
    

    View Slide

18. SYSTEM OVERVIEW
    

    View Slide

19. SYSTEM OVERVIEW
    

    View Slide

20. SYSTEM OVERVIEW
    

    View Slide

21. SYSTEM OVERVIEW
    

    View Slide

22. SYSTEM OVERVIEW
    

    View Slide

23. SYSTEM OVERVIEW
    

    View Slide

24. SYSTEM OVERVIEW
    

    View Slide

25. NATURE OF CENTRALIZED FEATURES
    Versatile Flexible Reusable Extensible
    

    View Slide

26. NATURE OF CENTRALIZED FEATURES
    Versatile Flexible Reusable Extensible
    

    View Slide

27. NATURE OF CENTRALIZED FEATURES
    Versatile Flexible Reusable Extensible
    

    View Slide

28. NATURE OF CENTRALIZED FEATURES
    Versatile Flexible Reusable Extensible
    

    View Slide

29. NATURE OF CENTRALIZED FEATURES
    Versatile Flexible Reusable Extensible
    

    View Slide

30. Available Features
    

    View Slide

31. AVAILABLE FEATURES
    Z-Features
    User

    Features
    Y-Features
    Obfuscated 

    User Features
    C-Features
    Content
    Features
    

    View Slide

32. Z-FEATURES
    BACKGROUND
    GENDER
    AGE-
    GROUP REGION
    User Demographics

    Estimation
    Look-a-like

    Engine
    

    View Slide

33. Z-FEATURES
    BACKGROUND
    Input
    Sparse vector

    from

    user's behaviral logs
    Output
    Class probabilities
    

    View Slide

34. > Collection of users' behavioral logs across various LINE services
    Z-FEATURES
    OVERVIEW
    

    View Slide

35. > Collection of users' behavioral logs across various LINE services
    Z-FEATURES
    OVERVIEW
    Transform into structures 

    that cover about 80% of 

    all ML use cases
    

    View Slide

36. > Collection of users' behavioral logs across various LINE services
    Z-FEATURES
    OVERVIEW
    {...} {...}
    {...} {...}
    {...} {...}
    ...
    ...
    

    View Slide

37. Z-FEATURES
    STATISTICS
    Dimensions
    50M+
    Users
    890M+
    Types
    30+
    Services
    10+
    

    View Slide

38. Z-FEATURES
    COMPONENTS USING Z-FEATURES
    

    View Slide

39. AVAILABLE FEATURES
    Z-Features
    User

    Features
    Y-Features
    Obfuscated 

    User Features
    C-Features
    Content
    Features
    

    View Slide

40. Y-FEATURES
    BACKGROUND
    Human-
    interpretable
    Extremely
    sparse
    

    View Slide

41. > Obfuscated user features
    > Mitigate z-features' problems
    • Accumulate content embedding based on users' behavioral logs
    • Reduce dimensionality
    Y-FEATURES
    OVERVIEW
    

    View Slide

42. > Obfuscated user features
    > Mitigate z-features' problems
    • Accumulate content embedding based on users' behavioral logs
    • Reduce dimensionality
    Y-FEATURES
    OVERVIEW
    

    View Slide

43. > Obfuscated user features
    > Mitigate z-features' problems
    • Accumulate content embedding based on users' behavioral logs
    • Reduce dimensionality
    Y-FEATURES
    OVERVIEW
    

    View Slide

44. > Obfuscated user features
    > Mitigate z-features' problems
    • Accumulate content embedding based on users' behavioral logs
    • Reduce dimensionality
    Y-FEATURES
    OVERVIEW
    Matrix sketching + PCA
    

    View Slide

45. Y-FEATURES
    STATISTICS
    Dimensions
    60K
    Users
    400M+
    Types
    20+
    Services
    10+
    

    View Slide

46. Y-FEATURES
    USER DEMOGRAPHICS ESTIMATION FOR JP REGION
    GENDER ESTIMATION METRICS

    (RELATIVE TO Z-FEATURES)
    0
    0.25
    0.5
    0.75
    1
    precision recall f1-score
    1.00
    1.00
    0.99
    AGE-GROUP ESTIMATION METRICS

    (RELATIVE TO Z-FEATURES)
    0
    0.25
    0.5
    0.75
    1
    precision recall f1-score
    0.88
    0.88
    0.88
    REGION ESTIMATION METRICS

    (RELATIVE TO Z-FEATURES)
    0
    0.25
    0.5
    0.75
    1
    precision recall f1-score
    0.98
    0.98
    0.99
    

    View Slide

47. Y-FEATURES
    USER DEMOGRAPHICS ESTIMATION FOR JP REGION
    TRAINING TIME

    (RELATIVE TO Z-FEATURES)
    0
    0.25
    0.5
    0.75
    1
    gender age-group region
    0.06
    0.02
    0.05
    PREDICTION TIME

    (RELATIVE TO Z-FEATURES)
    0
    0.25
    0.5
    0.75
    1
    gender age-group region
    0.52
    0.51
    0.20
    

    View Slide

48. Y-FEATURES
    COMPONENTS USING Y-FEATURES
    User to User

    Recommendation
    CTR/CVR Prediction

    on Ads Platform
    

    View Slide

49. FEATURES
    Z-Features
    User

    Features
    Y-Features
    Obfuscated 

    User Features
    C-Features
    Content
    Features
    

    View Slide

50. C-FEATURES
    OVERVIEW
    > Embedding of each service's contents
    > Currently available for two services
    • News articles: SCDV with fastText
    • Sticker images: Xception
    

    View Slide

51. C-FEATURES
    STATISTICS
    Dimension
    15K
    Contents
    3M+
    Types
    5
    Services
    2
    

    View Slide

52. Conclusion
    

    View Slide

53. HOW WE USE FEATURES AT DATA LABS
    > Feature as a Service
    • Achieve data standardization/democratization
    • Improve development efficiency
    > Available Features
    • User features
    • Obfuscated user features
    • Content features
    

    View Slide

54. Thank You
    

    View Slide