the rise of recommendation engines
TRANSCRIPT
The rise of Recommendation
Who is Gravity R&D?
Gravityuses
machine learningand
Big Data analyticsto create
personalized user journeysacross
all touch points
Scientific Exellence• 4 PhDs in the founding team since 2006• Tied for first place in the Netflix Prize• 200+ journal publications that have been cited over 2,000
times• Patented algorithms USPTO 8,676,736
What is Big Data
Information overloadEach day
500 million tweets 1 billion active user on Facebook 750,000 hours of video uploaded to Youtube
Not just about the size of data sets
Gartner: Big data is high volume, high velocity, and/or high variety information assets that require new forms of processing to enable enhanced decision making, insight discovery and process optimization.
What do we do with a lot of data?
Portal
Search Engine
Information aggregation
Information retrieval
Recommendationsystem
Information filtering
Recommendation systems …
… help users find content that is relevant to their interests
Gravity R&D
Examples
Netflix
Amazon
Tiki.vn
35% revenue come from recommended products
75% of what people watch is from personalization
Why Recommendation?
It’s not information overload.It’s filter failure
Clay Shirky
interesting book or cool music
What are the right keywords?
Paradox of Choice
Fewer Options lead to More Actions
More Options lead to Fewer Actions
A lot of times, people don’t know what they want until you show it
to themSteve Jobs
Recommendation in Vietnam
Challenges• Completely from scratch
Scientific knowledge Implementation algorithms into code
• Open source solution Research about the solution System architecture Fine tuning Maintenance
• Technology vendor You have to pay
ChallengeHow to keep users engaged and generate extra margin on existing users
SolutionPersonalization on the Home Page and Similar Item on Product and Cart Page. Cross recommendation on mobile app and email.
ResultsGravity significantly outperformed both other competitors in AB test.
Gravity’s recommendations resulted in...
“Gravity has been outperforming Strands and TargetingMantra in every important aspect and KPI: Revenue, AOV, CTR and response time.” — Hung Tran Viet, Product Owner, Tiki.vn
a GMV of
$13.15
/1000 recs
reached
6 %conversion
rate
ChallengeIncrease the discoverability of jobs and the relevant of job offers in the newsletter
SolutionGravity show similar jobs on the job detail page and personalize the jobs listed in the newsletter based on user behavior.
ResultsJob applicants are much more engaged with the personalized list of jobs in compare to the vanilla list
Users viewed Gravity’s recommendation …
“Gravity provides a job recommendation solution that has plug and play integration and works out of the box. Our new job recommendation engine helps our jobseekers discover their dream jobs. ” — Eduardo Mora, Head of Product & Engineering, Vietnamworks
clicks
59%more often
apply
26%more
The Future
• ML theories often based on long logs of user preference• In practice sometimes not feasible• People use item2item methods instead
relatively good result cheap to implement/calculation but only use last click information for recommendation
• DL models user’s session preference first click as input sequential input feeded into DL layers
• Thumbnail image as input
Deep Learning based Recommendation
• Forrester Research: more than $1 trillion of retail sales in 2015 were influenced by mobile phones• Personalized promotion pushed via app notification when
customers visit store• Follow-up email if customer leaves store without purchase• Suggestion for call center agents
O2O – Omnichannel
Right Product - Right Price - Right Channel - Right Time
O2O – Omnichannel
• Self-servicing for SME• Quick and simple integration• Plugins with major ecommerce platforms Shopify, Magento,
Haravan
Recommendation as a Platform
Thank you!
Contact us:
For the latest trends and insights: www.facebook.com/gravityrd