Slide 1
Slide 1 text
ANONYMIZE LARGE-SCALE
SPARSE USER FEATURES
AT LINE CORP
CHAERIM YEO, LINE CORPORATION
MACHINE LEARNING PRODUCTION PITCH #1, 2019/03/07
Slide 2
Slide 2 text
ABOUT ME
l Chaerim Yeo(呂 彩林)
l 2018.12 ~ LINE Corporation
l Account Platform Development Dept.
l Ad performance optimization
Slide 3
Slide 3 text
Agenda • Z-Features
• Y-Features
• Evaluation
• Conclusion
Slide 4
Slide 4 text
Z-FEATURES
Slide 5
Slide 5 text
WHAT ARE Z-FEATURES
Slide 6
Slide 6 text
WHAT ARE Z-FEATURES
Slide 7
Slide 7 text
WHAT ARE Z-FEATURES
Slide 8
Slide 8 text
WHAT ARE Z-FEATURES
Slide 9
Slide 9 text
WHAT ARE Z-FEATURES
Slide 10
Slide 10 text
BENEFIT OF Z-FEATURES
Reusable Flexible
Slide 11
Slide 11 text
LIMITATION OF Z-FEATURES
Human
Interpretable
Extremely
Sparse
Slide 12
Slide 12 text
Y-FEATURES
Slide 13
Slide 13 text
BEYOND Z-FEATURES
Obfuscation
Dimensionality
Reduction
Slide 14
Slide 14 text
BEYOND Z-FEATURES
Obfuscation
Dimensionality
Reduction
With keeping
information
as far as possible
Slide 15
Slide 15 text
BEYOND Z-FEATURES
Obfuscation
Dimensionality
Reduction
SCDV
https://arxiv.org/abs/1612.06778
Slide 16
Slide 16 text
OVERVIEW OF SCDV
Slide 17
Slide 17 text
INTEGRATE Z-FEATURES WITH SCDV
Slide 18
Slide 18 text
SYSTEM OVERVIEW
Slide 19
Slide 19 text
EVALUATION
Slide 20
Slide 20 text
DATA DIMENSION
RELATIVE TO Z-FEATURES (LOG-SCALE)
0.0001
0.0010
0.0100
0.1000
1.0000
10.0000
100.0000
type1 type2 type3 type4 type5 type6 type7 type8 type9
Slide 21
Slide 21 text
DATA DENSITY
LOG-SCALE
0.0000001
0.0000010
0.0000100
0.0001000
0.0010000
0.0100000
0.1000000
1.0000000
type1 type2 type3 type4 type5 type6 type7 type8 type9
z-features
y-features
Slide 22
Slide 22 text
DATA SIZE
RELATIVE TO Z-FEATURES
0.00
5.00
10.00
15.00
20.00
25.00
30.00
35.00
40.00
45.00
50.00
type1 type2 type3 type4 type5 type6 type7 type8 type9
Slide 23
Slide 23 text
USER DEMOGRAPHICS ESTIMATION
MATRICS (RELATIVE TO Z-FEATURES)
0.95
0.96
0.97
0.98
0.99
1.00
1.01
1.02
gender age-group region
precision recall f1-score
Slide 24
Slide 24 text
USER DEMOGRAPHICS ESTIMATION
RUNNING TIME (RELATIVE TO Z-FEATURES)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.50
gender age-group region
training prediction
Slide 25
Slide 25 text
CONCLUSION
Slide 26
Slide 26 text
CONCLUSION
l Anonymize user features based on SCDV
l Enough to use in ML
l Future works
l Add workflow to production
l Apply further dimensionality reduction
l Auto encoders, PCA, …
Slide 27
Slide 27 text
THANK YOU