Changing Advertising in China Bessie Lee Ching - - PowerPoint PPT Presentation

How Artificial Intelligence is Changing Advertising in China

Bessie Lee Ching Law

X

Founder & CEO - Withinlink General Manager - Tencent Social Ads

About Tencent

X

Tencent was founded

1998.11

1999.02 Launched instant messaging service- QQ Entered online games market 2003.08 Launched web portal – www.QQ.com 2003.12

Listed on Hong Kong Stock Exchange

2004.06

Launched social networking service Qzone 2005.05 Established Tencent Charity Fund 2007.06 Annual revenue

• ver RMB 10

billion 2009.12

Launched Weixin

2011.01

Launched open platform strategy 2011.06 Mobile QQ and Weixin launched game centers 2013.08 Market capitalization

• ver US$ 100

billion 2013.09

"Internet+" concept proposed by Pony Ma was included in the National Strategy Report

2015.01

Market capitalization

• ver US$ 300

billion 2017.05 Fortune Global 500 Companies 2017 Tencent included

Tencent Has Developed into A Leading Provider of Internet Services in China

X

6 Business Groups Provide A Range of Internet Products

X

CSIG

Cloud & Smart Industries Group

WXG

W eixin Group

IEG

Interactive Entertainment Group

PCG

Platform & Content Group

TEG

Technical Engineering Group

CDG

Corporate Development Group Tencent Cloud Qzone

Wechat

WeChat Work QQ Mail

Tencent Games

Tencent animation&comics China Literature Tencent Pictures Tencent eSports QQ Browser YingYongBao PC Manager Mobile Manager Tencent Auto lntelligence Tencent Internet Plus QQ Map Tencent Video Tencent Open Platform Tencent Big Data Tencent Data Center Security Financial Tech Tencent W iFi Manager Tencent News

Advertising & Marketing Service

Connecting People with People Connecting People with Services Connecting People with Devices

“Connection” Strategy

X

AI @ Tencent

X

Reinforcement Learning

Computer Vision Machine Learning Voice Recognition Natural Language Processing

Decision Making Generation Creation Cognition Comprehension

AI In All

X

Tencent Al Labs Wechat AI AI Lab Tencent Youtu

AI.QQ.COM

X

CASE 1 Smart Retail

Face Payment

Face recognition payment solution that empowers the retail industry

应用场景

Self-service cash register Small screen cash register Mobile self-service cash register Extremely efficient

1s

Precise identification

99%

Safe and reliable

3D Live detection

X

CASE 2 Medical Imaging

Locate suspicious nodules to improve the discovery rate and accuracy of early lung cancer It has been applied to more than 60 hospitals

Pulmonary nodule detection Lung cancer diagnosis AI electronic diagnosis report Lung cancer monitoring

X

AI in Tencent Ads

X

Ads Serving – Machine Gun for a Moving Target

X

Ad request User profile Best Ad

AD3 AD1 AD2

• Tens of billions of requests
• Hundreds of thousands of ads
• Real-time matching of the most relevant

ads to users

• Supported by machine learning models

trained on extremely large historical data

Intelligent Profile Understanding & Accurate User Insights

X

Demographics

• inherit
• location
• education
• marriage
• wealth
• employment

Device

• carrier
• network
• brand
• type
• OS
• price

Custom

• seed
• expansion
• 1-party labels

Interest

• purchasing

intent

• hobby

Vertical

• auto
• education
• eCommerce
• game

Behavior

• travel
• App
• eCommerce
• O2O
• Ad engagement
• purchase
• travel
• 3C
• finance
• FMCG

UID

Traditional User Definition - Manual Targeting

X

Advertisers manually specify the targeting criteria Large number of targetable attributes Boolean match during ads retrieval

Gender, age, geo, education. Marriage, employment, wealth, culture APP, eCommerce, Search, and Social behaviors Context, device, weather Interests, keywords Followers, APP installs, payments, engagements

Cons

• Prior knowledge
• Information loss
• Coarse graininess
• Manual optimization

True False False

Gender: F Age 30~60 Exclude: previous buyers

Interest: designer purse

Auto Targeting: Automatic User Identification by AI

X

A change of mindset

• Targeting by user-ad relevance

Targeting by machine learning models Multiple ways to define seed user Matching based on relevance Existing leads from 1st-party data Previously Converted users User/Ad features

0.81 0.79 0.56

x 1 x 2 x 3 +1 +1

Auto-Targeting：Representation Learning & ANN

X

Approximate Nearest Neighbor Search

Efficient ad retrieval based on user/ad embeddings under scalability constraints

”Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs ”, 2016

Representation learning

Learns embeddings for users and ads; can be used during ad retrieval Ad embedding User embedding

User features Context features Channel features Ad features

Lookalike Expansion for Targeting: Identify Potential New Customers

X

MORE

Understand the commonality of the seed audience Find lookalike audience

Lookalike

Custom audience based on known converters

Tencent data 1st-party data

User Profiles

Mined user features

Peacock

• Large scale distributed LDA

Logistic Regression

• One model per seed audience,

expansion done offline DNN

• Deep embedding for each seed user,

expansion done online 2013 2014 2015

Static Creative is Hard to Optimize for Advertisers with Large Inventory

X

Creation Serving Impression

Advertiser

User

Creative

 Hit-or-miss  Coarse-grained targeting  Static content with no personalization Challenges

Static Creative：

DPA: Automatic Creation, Retrieval, and Rendering of Personalized Ad Creatives

X

Product targeting Product retrieval Online creative generation Product DB Creation Serving Impression

Advertiser

User  Truly personalized

• Different experience even for

the same ad

 Automatic creative generation

• No need for trial-and-error
• ptimization from advertisers

 Fine-grained targeting

• 100m targetable products
• More detailed and accurate

than ad targeting

Advantages

Creative

AI Innovations by Chinese Startups

X

Collaboration AI Backend

X

Collaboration Marketing Solution

X

Collaboration Offline Data Platform

X

Collaboration KOL Management

X

Thank you

X

Contact Us

X

Bessie Lee

Bessie.lee@withinlink.com chinglaw@tencent.com

Ching Law