4. 4
Location & members
Strategic R&D organization in Rakuten Group
+150 researchers all over the world
Japan USSingapore
France
India
5. 5
Research Area of RIT
3 research groups for adapting to internet growth
RealityIntelligencePower
• HCI
• MR / AR
• Image Processing
• Robotics
• HPC
• Distributed Computing
• Programing Language
• Machine Learning
• Deep Learning
• NLP
• Data Mining
6. 6
Research Area of RIT
3 research groups for adapting to internet growth
RealityIntelligencePower
• HCI
• MR / AR
• Image Processing
• Robotics
• HPC
• Distributed Computing
• Programing Language
• Machine Learning
• Deep Learning
• NLP
• Data Mining
7. 7
Mission of Intelligence Domain Group
Assist Rakuten businesses to boost with ML and DM
Optimizing A/B testing
Item Classification
User Segmentation
Data + AI
Coupon Distribution
Recommender System Economy Prediction / Demand Prediction
Review Analysis
Anomaly Detection / Fraud Detection
Product Data Analysis
Business
9. 9
Mission of Intelligence Domain Group
Assist Rakuten businesses to boost with ML and DM
Optimizing A/B testing
Item Classification
User Segmentation
Data + AI
Coupon Distribution
Recommender System Economy Prediction / Demand Prediction
Review Analysis
Anomaly Detection / Fraud Detection
Product Data Analysis
Business
12. 12
Model
Input
Interest rate
Predict who will use particular financial service
• Purchase trend
• Purchase of each L2 genre
• GMS trend, Frequency
• Basic Demogra.
• Gender, Area, Age
Model
Output
Interest rate
Will be user
Won’t be user
13. 13
What kinds of Rakuten ICHIBA users
are most likely to apply for the target service?
15. 15
Evaluation #1
Prospective Users Control Group
• Randomly Selected
• About 300,000 users
• Score >= 0.8
• About 300,000 users
Send ichiba mail magazine to two groups
Ichiba Mail Magazine
16. 16
Evaluation #2
Mail Deliver
Open Mail
Click Contents
(Visit Service
Page)
Click Rate went up by +49.23%
compared with control group
+3.52% +49.23%
17. 17
Application in other services
Prospective
users
Extract
Target
service
Same scheme can be applied to various kinds of services
Ichiba Active Users
18. 18
Platform-nization
Algorithm: RIT
Platform: Data Science Dep.
Rakuten Businesses National Clients
Customer
DNA
• Input: Users who uses the particular service
• Output: Users who will use that service
19. 19
Platform-nization
Algorithm / Model : RIT
Platform / Interface : DSD
Rakuten Businesses National Clients
Customer
DNA
• Creating a platform for the solution together with GDSD.
• Integrating Customer DNA for User Features
23. 23
Mission of Intelligence Domain Group
Assist Rakuten businesses to boost with ML and DM
Optimizing A/B testing
Item Classification
User Segmentation
Data + AI
Coupon Distribution
Recommender System Economy Prediction / Demand Prediction
Review Analysis
Anomaly Detection / Fraud Detection
Product Data Analysis
Business
27. 27
Model
• Predict Composite Index by using LASSO
• Use monthly sales data in each L4 genre
Effect
Sales at t (genre a)
Composite Index
at t
(t 期の景気動向指数)
LASSO
Sales at t (genre b)
Sales at t (genre c)
:
L4 genres, 2521
genres
28. 28
Prediction Results
Mean absolute error is about 0.4
• Training data : Dec., 2009 – Dec. 2012
• Test data : Jan., 2013 – Apr., 2013
94
96
98
100
102
104
106
108
110
CompositeIndex(CI)
Month
Prediction of Composite Index
Actual Predict(Training fit)
Predict(Test fit)
32. 32
Mission of Intelligence Domain Group
Assist Rakuten businesses to boost with ML and DM
Optimizing A/B testing
Item Classification
User Segmentation
Data + AI
Coupon Distribution
Recommender System Economy Prediction / Demand Prediction
Review Analysis
Anomaly Detection / Fraud Detection
Product Data Analysis
Business
33. 33
See you later in the round table discussion!
Data
• Contribution for existing
business
• New Business Dev.
AI