Download - Traffic Shaping to Optimize Ad Delivery
![Page 1: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/1.jpg)
1
Traffic Shaping to Optimize Ad Delivery
Deepayan ChakrabartiErik Vee
![Page 2: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/2.jpg)
Traffic Shaping
2
Which article summary should
be picked?
Ans: The one with highest expected CTR
Which ad should be displayed?
Ans: The ad that minimizes underdelivery
Article pool
![Page 3: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/3.jpg)
Underdelivery
Advertisers are guaranteed some impressions (say, 1M) over some time (say, 2 months) only to users matching their specs only when they visit certain types of pages only on certain positions on the page
An underdelivering ad is one that is likely to miss its guarantee
3
![Page 4: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/4.jpg)
Traffic Shaping
4
Which article summary should
be picked?
Ans: The one with highest expected CTR
Which ad should be displayed?
Ans: The ad that minimizes underdelivery
Goal: Combine the two
![Page 5: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/5.jpg)
Traffic Shaping
Goal: Bias the article summary selection to reduce under-delivery but insignificant drop in CTR AND do this in real-time
![Page 6: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/6.jpg)
Outline
Formulation as an optimization problem Real-time solution Empirical results
6
![Page 7: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/7.jpg)
Formulation
j:(ads)
ℓ:(user, article, position)“Fully Qualified Impression”
i:(user, article)
k:(user)
ℓj
i
k
Goal: Infer traffic shaping fractions wki
Supply sk
CTR c ki
Traffic
shaping
fractio
n w ki
Demand dj
Ad delivery fraction φℓj
![Page 8: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/8.jpg)
Formulation
Full traffic shaping graph: All forecasted user traffic X
all available articles arriving at the homepage, or directly on article page
Goal: Infer wki But forced to infer φℓj as
wellFull Traffic Shaping Graph
A
B
C
Traffic
shaping
fractio
n w ki
Ad delivery fraction φℓj
CTR c ki
![Page 9: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/9.jpg)
Outline
Formulation as an optimization problem Real-time solution Empirical results
9
![Page 10: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/10.jpg)
Formulation Reformulation: {wki, φℓj}→ zℓj
Convex program can be solved optimally
10
![Page 11: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/11.jpg)
Formulation
But we have another problem At runtime, we must shape every incoming user
without looking at the entire graph
Solution: Periodically solve the convex problem offline Store a cache derived from this solution Reconstruct the optimal solution for each user at
runtime, using only the cache
11
![Page 12: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/12.jpg)
Real-time solution
12
Cache these
Reconstruct using these
All constraints can be expressed as constraints on σℓ
![Page 13: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/13.jpg)
Results
Data: Historical traffic logs from April, 2011 25K user nodes
Total supply weight > 50B impressions 100K ads
13
![Page 14: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/14.jpg)
Lift in impressionsLi
ft in
impr
essi
ons
deliv
ered
to
unde
rper
form
ing
ads
Fraction of traffic that is not shaped
Nearly threefold improvement via
traffic shaping
14
![Page 15: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/15.jpg)
Average CTRA
vera
ge C
TR (a
s pe
rcen
tage
of
max
imum
CTR
)
Fraction of traffic that is not shaped
CTR drop < 10%
15
![Page 16: Traffic Shaping to Optimize Ad Delivery](https://reader035.vdocument.in/reader035/viewer/2022081520/5681662f550346895dd9940d/html5/thumbnails/16.jpg)
Summary
3x underdelivery reduction with <10% CTR drop 2.6x reduction with 4% CTR drop Runtime application needs only a small cache
17