boosting consumer engagement at paypal
TRANSCRIPT
Boosting Consumer Engagement
at PayPal
Sujit Mathew, Goh Yew Yap, Chen Yanhui
PayPal
Two decades ago, our founders invented payment technology to make buying and selling faster, safer, and easier—and put economic power where it belongs: In the hands of people.
Mass Adoption of
Mobile Devices
Digitization
of Cash
Transformation
of Cards
Fragmentation of
Payment Types,
Technology and
Channels
Rise of Fraud
and Cybercrime
Money is
changing
PayPal is leading the transformation
AT SCALE*
173 Million Customers**
$235 Billion TPV
$8 Billion Revenue
4 Billion Transactions
WITH MOMENTUM
+19 Million Customers Gained in 2014
+26% y/y TPV Growth
+22% y/y Transaction Growth
©2015 PayPal Inc. Confidential and proprietary.
PayPal processed $46 billion in mobile payment volume in 2014,
up 68% over 2013.
In 2014, 20% of PayPal’s net Total Payment Volume was
from mobile payments.
Venmo processed $2.4B in Total Payment Volume in 2014.
In Q3 2015 Venmo’s Payment Volume was $2.11 billion – up 201% year over year.
In 2014, PayPal and Venmo combined handled billions in P2P payment volume globally and nearly half of that volume was international.
A leader in
person-person payments
Use Case
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary. 8
“Boost consumer engagement by
recommending merchants and products
to consumers.”
©2015 PayPal Inc. Confidential and proprietary.
66 Million individual payments processed for charities via PayPal.
36 Million consumers used PayPal to make a payment to a charity.
13% of donations through PayPal in 2014 were made on a mobile device.
418,000 charities used PayPal to accept donations.
$5.7 Billion processed for charities by PayPal.
65% YoY growth in PayPal’s total mobile payment value to charities globally.
Bridging Consumers and Charities
Modeling
Overview
Stack
Graph
Collaborative Filtering
Content Model
Deployment
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Technology Stack
12
Hadoop / MapReduce
Mahout Pig
Python / Shell
HDFS Cassandra
Titan
Gremlin
Graph Modeling
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Build Property Graph Based on P2P transaction data.
14
Discover Communities within P2P
Data
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Discover Key Influencers
15
Eigenvector Centrality
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Property Graphs
G = (V , E , λ)
V = vertices
E = Edges
λ = Properties
16
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Property Graphs
17
User Type
Age
Influence Score
Charity Type
Name
g.V[[type:"Charity"]]
.inE("Donate")
.filter{it.getProperty('Amount') > 25}
.outV.filter{it.getProperty('Influence') >
0.5}
Amount
Recommend
Enrich the graph with Donations and Social data.
Key Interests within
group
Charity
Send
Find all Key influencers who
have donated more than 25
USD to the charity
Collaborative Filtering
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Collaborative Filtering 101
19
Commerce Interaction Matrix
We want to model the affinity
between consumers and merchants
More transactions occurred,
more confident we believe the relationship
Consumer
nonprofits
Likeness
Matrix
Confidence
Matrix
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Collaborative Filtering 101 A matrix factorization method
20
Data Fitting Regularization
Merchant
nonprofits
Consum
er
Consum
er
d
d
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Alternative Least Square
21
Iteratively update
Fix V and update U:
Fix U and update V:
Regularization Data Fitting
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Scalable Collaborative Filtering Improve the scalability of ALS
22
Shard 1 Shard 2 Shard 3 Consumer
Merchant
Commerce Interaction Matrix
Mapreduce Job
for Shard 1
Mapreduce Job
for Shard 2
Mapreduce Job
for Shard 3
Stage 1:
Compute the individual
contributions of each rating
Stage 2:
Aggregate all contributions for every
user and update their models in
parallel
Reference: http://www.slideshare.net/jekky_yiqun/scalable-collaborative-filtering-for-commerce-recommendation
Global MapReduce Job for ALS
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
What does CF Learn?
23
03.05.0
15.09.0
2.11.0
3.001.001.03.12.07.0
VU T
Each vector of U model a consumer
by d implicit attributes
Each vector of V model a merchant
by d implicit attributes
Score = UiT . Vj = 1.162
Content Model
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Data & Feature Dictionary
cust_id
string
age
range
[10:20:30:40:50:]
gender
categorical
[M,F,U]
country
categorical
[NP, IN, CN]
spend
numeric
nonprofits
label
Cust ID Age … Gender Countr
y Spend nonprofits
1 28 M NP 10.5 1
2 35 F CN 100 2
3 30 M IN 25.1 1
4 34 F IN 15 3
5 32 M IN 5 4
6 25 F IN 22.5 1
3 30 M IN 12 3
3 30 M IN 1 2
Feature Dictionary Dataset
Name
Type
Values
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Training the model
Data
Source
Featurizer
Business
Logic
Feature
Dictionary Other
resources
Features Learner Predictive
Models
ML Algorithm
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Logistic Regression
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Recurrent Neural Network - LSTM
Source: http://colah.github.io/posts/2015-08-Understanding-LSTMs/
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Combining the result
Teradata
& HDFS
Interaction
Matrix
Features
CF
Model
Content
Model
Hybrid
Result
CF Engine
Ensemble
…
Deployment
• Emails, Web, Mobile
Applications
• REST APIs Platform
• ML Models Data
Science
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
REST API
The
recommendations will
be stored.
MODEL
The
recommendations
will be generated
and uploaded.
FRONT-END WEB
will query for
recommendations
Recommendations
of respective
customer will be
returned.
Web
Emails
Mobile
Offers
© 2015 PayPal Inc. All rights reserved. Confidential and proprietary.
Wrap Up
• Approaches to modeling
– Graph Model
– CF Model
– Content Model
• Deployment of models
Thank You
Sujit Mathew, Goh Yew Yap, Chen Yanhui