advertising engines - a guide to web research: lecture 1 › webguide › 01webguide.pdf · age,...
TRANSCRIPT
![Page 1: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/1.jpg)
Advertising EnginesA Guide to Web Research: Lecture 1
Yury Lifshits
Steklov Institute of Mathematics at St.Petersburg
Stuttgart, Spring 2007
1 / 28
![Page 2: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/2.jpg)
new algorithmicproblems
new modelsand notions
2 / 28
![Page 3: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/3.jpg)
Talk Objective
Industrial solutions
Google AdWords
Google AdSense
Yahoo! SearchMarketing
Microsoft adCenter
Amazon recommendations
Coming soon: personalized ads for webmail, social networks,blogging platforms, phones, computer games, supermarket billsetc.
Today we show(1) single model for distributing personalized ads(2) open algorithmic problems motivated by such systems
3 / 28
![Page 4: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/4.jpg)
Talk Objective
Industrial solutions
Google AdWords
Google AdSense
Yahoo! SearchMarketing
Microsoft adCenter
Amazon recommendations
Coming soon: personalized ads for webmail, social networks,blogging platforms, phones, computer games, supermarket billsetc.
Today we show(1) single model for distributing personalized ads(2) open algorithmic problems motivated by such systems
3 / 28
![Page 5: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/5.jpg)
Outline
1 Architecture of Advertising EnginesComponent 1: EventComponent 2: AdvertiserComponent 3: Advertising Engine
2 Algorithmic ChallengesTarget optimizationClick VolumeAdRank ComputingAd Coverings
4 / 28
![Page 6: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/6.jpg)
Outline
1 Architecture of Advertising EnginesComponent 1: EventComponent 2: AdvertiserComponent 3: Advertising Engine
2 Algorithmic ChallengesTarget optimizationClick VolumeAdRank ComputingAd Coverings
4 / 28
![Page 7: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/7.jpg)
Part I: Architecture of Advertising Engines
5 / 28
![Page 8: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/8.jpg)
Example: Sponsored Search
Stuttgart houseSearchEngine
1
2
3
Advertisement Advertisement
Advertisement
Advertisement
Algorithmic
results
6 / 28
![Page 9: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/9.jpg)
Example: Context Ads
Stuttgart Estate AgencyAdvertisement
Advertisement
Advertisement
Main
content
7 / 28
![Page 10: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/10.jpg)
Three Components: Event, Advertiser, Engine
Event
PersonMediaAction
Advertiser
Ad$$$Targeting
AdvertisingEngine
Choosingprocedure
Pricingmechanism
FOR SALEwww.home.org
8 / 28
![Page 11: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/11.jpg)
Component 1: Event
Collect all available information:
Person: What do we know about him/her?
Age, geographic location, previous actions, interests etc
Media: What is situated around the ad placement?
Content and typical audience of website, tv program, newspaper
Action: Current relations between person and media?
Current search query, purchasing a book, signing up to a service
9 / 28
![Page 12: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/12.jpg)
Component 2: Advertiser
FOR SALEwww.home.org
Setting new campaign:Ad: What will be displayed?
Text, image, video, hyperlink, phone number, advertiser’swebsite
$$$: Size of campaign?Monthly/daily budget, maximal admissible price (bid) forclick/impression
Targeting: Who is target audience?Location, specific query keywords, category of landing page
Targeting in general: any subset of event space P × M × A10 / 28
![Page 13: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/13.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 14: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/14.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 15: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/15.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 16: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/16.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 17: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/17.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 18: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/18.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 19: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/19.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 20: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/20.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 21: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/21.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 22: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/22.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 23: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/23.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 24: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/24.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 25: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/25.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 26: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/26.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 27: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/27.jpg)
Component 3: Advertising Engine
Basic routine of advertising engine:
1 Get all available info about current event
2 Keep only ads that include this event totheir target
3 Rank ads according to their bids andtheir relevance to the event
4 Display (several) best ones
5 In case of click compute discount (actualprice for advertiser)
Events Ads
11 / 28
![Page 28: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/28.jpg)
Objectives
User
Maintain privacyReceive only relevant ads
Advertiser
Cheap clicksGet “relevant” clicks (high conversion rate)Transparent pricing and targeting mechanisms
Advertising Engine
Organize enough relevant clicks for any budget and any targetKeep prices highKeep users/advertisers happy
More objectives?
12 / 28
![Page 29: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/29.jpg)
Objectives
User
Maintain privacyReceive only relevant ads
Advertiser
Cheap clicksGet “relevant” clicks (high conversion rate)Transparent pricing and targeting mechanisms
Advertising Engine
Organize enough relevant clicks for any budget and any targetKeep prices highKeep users/advertisers happy
More objectives?
12 / 28
![Page 30: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/30.jpg)
Objectives
User
Maintain privacyReceive only relevant ads
Advertiser
Cheap clicksGet “relevant” clicks (high conversion rate)Transparent pricing and targeting mechanisms
Advertising Engine
Organize enough relevant clicks for any budget and any targetKeep prices highKeep users/advertisers happy
More objectives?
12 / 28
![Page 31: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/31.jpg)
Objectives
User
Maintain privacyReceive only relevant ads
Advertiser
Cheap clicksGet “relevant” clicks (high conversion rate)Transparent pricing and targeting mechanisms
Advertising Engine
Organize enough relevant clicks for any budget and any targetKeep prices highKeep users/advertisers happy
More objectives?
12 / 28
![Page 32: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/32.jpg)
Part II: Algorithmic Challenges
Disclaimer: my style is
1 At first, think independently (e.g. pose new problems)
2 Only after that look into literature
Hence, the following problems might be already known and heavilystudied!
13 / 28
![Page 33: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/33.jpg)
Part II: Algorithmic Challenges
Disclaimer: my style is
1 At first, think independently (e.g. pose new problems)
2 Only after that look into literature
Hence, the following problems might be already known and heavilystudied!
13 / 28
![Page 34: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/34.jpg)
Target optimization (1/4) Informally
Advertiser sets target audience. Advertising engine should help:
Some potentially interested people are missed
Exclude people who will be offended by this ad
Proper setting of target audience is difficult
Advertising engine knows much more about event space
14 / 28
![Page 35: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/35.jpg)
Target optimization (2/4) Formalization
Events are vectors
Advertiser provides some sample events S = {e1, . . . , ek} from thetarget
Advertising engine produces an effective membership procedure foroptimized target S
How to define optimized target?
15 / 28
![Page 36: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/36.jpg)
Target optimization (2/4) Formalization
Events are vectors
Advertiser provides some sample events S = {e1, . . . , ek} from thetarget
Advertising engine produces an effective membership procedure foroptimized target S
How to define optimized target?
15 / 28
![Page 37: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/37.jpg)
Target optimization (3/4) Solution
Let B(e, r) be the ball in event space with center e and radius r
New target =⋃
k B(ek , r)
Effective membership for enew :
1 Find nearest representative ei ∈ S
2 Check whether Dist(enew , ei) < r
16 / 28
![Page 38: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/38.jpg)
Target optimization (3/4) Solution
Let B(e, r) be the ball in event space with center e and radius r
New target =⋃
k B(ek , r)
Effective membership for enew :
1 Find nearest representative ei ∈ S
2 Check whether Dist(enew , ei) < r
16 / 28
![Page 39: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/39.jpg)
Target optimization (3/4) Solution
Let B(e, r) be the ball in event space with center e and radius r
New target =⋃
k B(ek , r)
Effective membership for enew :
1 Find nearest representative ei ∈ S
2 Check whether Dist(enew , ei) < r
16 / 28
![Page 40: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/40.jpg)
Target optimization (3/4) Solution
Let B(e, r) be the ball in event space with center e and radius r
New target =⋃
k B(ek , r)
Effective membership for enew :
1 Find nearest representative ei ∈ S
2 Check whether Dist(enew , ei) < r
16 / 28
![Page 41: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/41.jpg)
Target optimization (3/4) Solution
Let B(e, r) be the ball in event space with center e and radius r
New target =⋃
k B(ek , r)
Effective membership for enew :
1 Find nearest representative ei ∈ S
2 Check whether Dist(enew , ei) < r
16 / 28
![Page 42: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/42.jpg)
Target optimization (3/4) Solution
Let B(e, r) be the ball in event space with center e and radius r
New target =⋃
k B(ek , r)
Effective membership for enew :
1 Find nearest representative ei ∈ S
2 Check whether Dist(enew , ei) < r
16 / 28
![Page 43: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/43.jpg)
Target optimization (4/4) Questions Around
Other definitions for optimized target?
Exploiting historical information for target optimization
Target construction based on advertisement content
17 / 28
![Page 44: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/44.jpg)
Click Volume (1/4) Informally
Assume we show the same ad at all events. Then average dailyamount of clicks is click volume for the given ad
Motivation
Understand how much can we sell
Evaluate the effectiveness of currentengines
The first step towards recognizinginterested audience
Use different strategies for(supply<demand) and (supply>demand)
click
click
click
18 / 28
![Page 45: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/45.jpg)
Click Volume (1/4) Informally
Assume we show the same ad at all events. Then average dailyamount of clicks is click volume for the given ad
Motivation
Understand how much can we sell
Evaluate the effectiveness of currentengines
The first step towards recognizinginterested audience
Use different strategies for(supply<demand) and (supply>demand)
click
click
click
18 / 28
![Page 46: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/46.jpg)
Click Volume (1/4) Informally
Assume we show the same ad at all events. Then average dailyamount of clicks is click volume for the given ad
Motivation
Understand how much can we sell
Evaluate the effectiveness of currentengines
The first step towards recognizinginterested audience
Use different strategies for(supply<demand) and (supply>demand)
click
click
click
18 / 28
![Page 47: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/47.jpg)
Click Volume (1/4) Informally
Assume we show the same ad at all events. Then average dailyamount of clicks is click volume for the given ad
Motivation
Understand how much can we sell
Evaluate the effectiveness of currentengines
The first step towards recognizinginterested audience
Use different strategies for(supply<demand) and (supply>demand)
click
click
click
18 / 28
![Page 48: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/48.jpg)
Click Volume (1/4) Informally
Assume we show the same ad at all events. Then average dailyamount of clicks is click volume for the given ad
Motivation
Understand how much can we sell
Evaluate the effectiveness of currentengines
The first step towards recognizinginterested audience
Use different strategies for(supply<demand) and (supply>demand)
click
click
click
18 / 28
![Page 49: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/49.jpg)
Click Volume (2/4) Basic Formula
Take daily history event-ad-?click:
(e1, a1, b1) . . . (en, an, bn)
Use similarity-between-ads function S for computing click volume V :
V (anew) =∑
S(anew , ai) · bi
Any comments/objections?
19 / 28
![Page 50: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/50.jpg)
Click Volume (2/4) Basic Formula
Take daily history event-ad-?click:
(e1, a1, b1) . . . (en, an, bn)
Use similarity-between-ads function S for computing click volume V :
V (anew) =∑
S(anew , ai) · bi
Any comments/objections?
19 / 28
![Page 51: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/51.jpg)
Click Volume (3/4) Corrected Formula
Problem: click volume is underestimated since not all chosen ads aresimilar to anew
First step: prediction of click-through rate for a given event-ad pair
CTR(e, anew) =
∑ei=e S(ai , anew)bi∑ei=e S(ai , anew)
Second step: using click rates
V (anew) =∑
1≤i≤n
CTR(ei , anew)
20 / 28
![Page 52: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/52.jpg)
Click Volume (3/4) Corrected Formula
Problem: click volume is underestimated since not all chosen ads aresimilar to anew
First step: prediction of click-through rate for a given event-ad pair
CTR(e, anew) =
∑ei=e S(ai , anew)bi∑ei=e S(ai , anew)
Second step: using click rates
V (anew) =∑
1≤i≤n
CTR(ei , anew)
20 / 28
![Page 53: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/53.jpg)
Click Volume (3/4) Corrected Formula
Problem: click volume is underestimated since not all chosen ads aresimilar to anew
First step: prediction of click-through rate for a given event-ad pair
CTR(e, anew) =
∑ei=e S(ai , anew)bi∑ei=e S(ai , anew)
Second step: using click rates
V (anew) =∑
1≤i≤n
CTR(ei , anew)
20 / 28
![Page 54: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/54.jpg)
Click Volume (4/4) Questions Around
Computing ad volume (the amount of advertisements that canget positive response at the given event)
Fast algorithm for predicting click volume for all ads in thesystem
Exploiting metric inside event space
21 / 28
![Page 55: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/55.jpg)
AdRank Computing (1/2) Informally
Input: event enew , set of all ads A. Choosing-ads principles:
Take the most content-relevant
Take the ones with best click-through rate
Take ads with maximal bids
AdRank(enew , a) = Bid(a) · (ContRel(enew , a) + CTR(enew , a))
22 / 28
![Page 56: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/56.jpg)
AdRank Computing (2/2) Questions Around
Actually, finding content-closest ads to the given input is just thenearest neighbor problem. We need:
Data structure for A for fast computing of best AdRank(enew , a)values
Accurate and fast prediction for CTR(enew , a)
All ads
Highest AdRank
23 / 28
![Page 57: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/57.jpg)
AdRank Computing (2/2) Questions Around
Actually, finding content-closest ads to the given input is just thenearest neighbor problem. We need:
Data structure for A for fast computing of best AdRank(enew , a)values
Accurate and fast prediction for CTR(enew , a)
All ads
Highest AdRank
23 / 28
![Page 58: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/58.jpg)
AdRank Computing (2/2) Questions Around
Actually, finding content-closest ads to the given input is just thenearest neighbor problem. We need:
Data structure for A for fast computing of best AdRank(enew , a)values
Accurate and fast prediction for CTR(enew , a)
All ads
Highest AdRank
23 / 28
![Page 59: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/59.jpg)
AdRank Computing (2/2) Questions Around
Actually, finding content-closest ads to the given input is just thenearest neighbor problem. We need:
Data structure for A for fast computing of best AdRank(enew , a)values
Accurate and fast prediction for CTR(enew , a)
All ads
Highest AdRank
23 / 28
![Page 60: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/60.jpg)
Ad Coverings Informally
Consider any publishers-subscribers graph (say, RSS feeds):
What is the minimal amount of placements to cover all (target)audience?
Given fixed amount of placements how many subscribers can wecover twice?
Publishers
Subscriberstarget group
minimal ad covering
24 / 28
![Page 61: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/61.jpg)
Ad Coverings Informally
Consider any publishers-subscribers graph (say, RSS feeds):
What is the minimal amount of placements to cover all (target)audience?
Given fixed amount of placements how many subscribers can wecover twice?
Publishers
Subscriberstarget group
minimal ad covering
24 / 28
![Page 62: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/62.jpg)
Ad Coverings Informally
Consider any publishers-subscribers graph (say, RSS feeds):
What is the minimal amount of placements to cover all (target)audience?
Given fixed amount of placements how many subscribers can wecover twice?
Publishers
Subscriberstarget group
minimal ad covering
24 / 28
![Page 63: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/63.jpg)
Other Directions in Advertising Engines
Optimal ad distribution in case when interested audience islarger than budget
Machine learning for advertising engines
Weighted targeting (some events are preferable to others)
Advertising engines for social networks
Auction design for sponsored search
Click fraud
25 / 28
![Page 64: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/64.jpg)
Call for participation
Know a relevant reference?
Have an idea?
Find a mistake?
Solved one of these problems?
Knock to my office 1.156
Write to me [email protected]
Join our informal discussions
Participate in writing roadmap-paper
26 / 28
![Page 65: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/65.jpg)
Summary
Three components:
FOR SALEwww.home.org
Four problems:
Vielen Dank fur Ihre Aufmerksamkeit! Fragen?
27 / 28
![Page 66: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/66.jpg)
Summary
Three components:
FOR SALEwww.home.org
Four problems:
Vielen Dank fur Ihre Aufmerksamkeit! Fragen?
27 / 28
![Page 67: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/67.jpg)
Summary
Three components:
FOR SALEwww.home.org
Four problems:
Vielen Dank fur Ihre Aufmerksamkeit! Fragen?27 / 28
![Page 68: Advertising Engines - A Guide to Web Research: Lecture 1 › webguide › 01webguide.pdf · Age, geographic location, previous actions, interests etc Media: What is situated around](https://reader033.vdocuments.site/reader033/viewer/2022042323/5f0d39267e708231d43947b3/html5/thumbnails/68.jpg)
Sources
Course homepage http://logic.pdmi.ras.ru/~yura/webguide.html
Daniel C. Fain and JanO. Pedersen
Sponsored Search: a Brief History
http://www.bus.ualberta.ca/kasdemir/ssa2/fain pedersen.PDF
Alexander Tuzhilin
The Lanes Gifts v. Google Report
http://googleblog.blogspot.com/pdf/Tuzhilin Report.pdf
Moira Regelson and Daniel C. Fain
Predicting ClickThrough Rate Using Keyword Clusters
http://www.bus.ualberta.ca/kasdemir/ssa2/regelson fain.pdf
Juan Feng, HemantK. Bhargava and DavidM. Pennock
Implementing Sponsored Search in Web Search Engines: Computational Evaluation ofAlternative Mechanisms
http://research.yahoo.com/node/338/2371
Panel Discussion at SSA2
Models for Sponsored Search: What are the right questions?
http://research.microsoft.com/~hartline/papers/panel-SSA-06.pdf28 / 28