Panagiotis Papadimitriou Hector GarciaMolina Stanford University Ali Dasdan Santanu Kolay Ebay Inc Related papers VLDB 2011 InfoLab TR939 AdAuctions 2009 Search Engine Results Page SERP ID: 243629
Download Presentation The PPT/PDF document "Output URL Bidding" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Output URL Bidding
Panagiotis Papadimitriou, Hector Garcia-Molina, (Stanford University)Ali Dasdan, Santanu Kolay(Ebay Inc)
Related papers: VLDB 2011,
InfoLab
TR-939,
AdAuctions
2009Slide2
Search Engine Results Page (SERP)
Organic Results
Sponsored Ads
Query
Sponsored Search AdsSlide3
Keyword Bidding
Advertiser
Search Engines
the social network
lord of the rings
the matrix
l
otr
III
...
...
# keywords = ~ 10K
KEYWORDSSlide4
Example SERPs
en.wikipedia.org
/wiki/
The_Social_Network
www.imdb.com
/title/tt1285016/
www.imdb.com
/title/tt133093/
en.wikipedia.org
/wiki/
The_Matrix
en.wikipedia.org
/wiki/
The_Lord_of_the_rings
en.wikipedia.org
/wiki/
The_Lord_of_the_rings
www.imdb.com
/title/tt167260/
www.imdb.com
/title/tt120737/
the social network
the matrix
the lord of the rings
lotr
iiiSlide5
Output Bidding
Advertiser
Search Engines
imdb.com AND wikipedia.org
# URLs = 2
URLsSlide6
Outline
Architectures Bid LanguageOutput bid/expression generationSpill EvaluationExperimentsSlide7
Architectures
Current Search Engine ArchitectureSlide8
Architectures
SerializationOverviewFirst, retrieve organic results Then, retrieve adsPros
SimplicityConsResults Latency
O: Organic Search System
S:
Sponsored Search System
SERPSlide9
Architectures
PipeliningSplit organic search system toOr: retrieval subsystem (retrieve relevant docs)Op: post-processing subsystem (create result snippets)Op and S run in parallel
ProsNo additional latencyConsSponsored search system depends on organic system
O:
Organic Search System =
Or
+
Op
S:
Sponsored Search System
SERPSlide10
Architectures
ParallelizationURLs with ads are known a prioriS can useOr’: Or replica that indexes only URLs with adsProsNo additional latencyIndependent organic and sponsored search system
ConsMore resources
O:
Organic Search System (
Or
+
Op
)
S:
Sponsored Search System
Or’: Small replica of OrV: Ad validation
SERPSlide11
Bid Language Model
Output Expressione.g., a := (u1 u2) u3 (h
1 h2
)u: URLe.g., en.wikipedia.org/wiki/The_Social_Network
h: host
e.g., en.wikipedia.org
Questions
URLs or hosts or both?
c
omplex or simple?Slide12
Output Expression Generation
MotivationUse existing keyword campaigns to generate realistic output expressions to study
The social network
lord of the ringsthe matrix
l
otr
III
…
…
Output Expression Generator
imdb.com AND wikipedia.org
Slide13
Problem
INPUT: keyword set ROUTPUT: expression a that “covers” RCandidate solutionsa1 := u1 u2 u
3a2 := u
1 u4a
3
:= u
5
Output Expression Generation
Motivating ExampleSlide14
Compactness
Contain few URLsSpill minimization:Do not match “irrelevant” queries
Output Expression Generation
Objectives
Output
Expression
Size
|a|
Spill
spill(
a,R
)
a
1 := u1 u2
u
3
3
{}
a
2
:= u
1
u4
2{q5}
a
3
:= u
5
1
{q
4
,q
5
, q
6
}Slide15
Query Set Output Cover
minimize γ|a| + (1-γ) |spill(a, R)|subj. to m(a,q
), q Rγ
: regularization parameterRelated toSet CoverRed-Blue Set Cover
Output Expression Generation
Problem StatementSlide16
Output Expression Generation
Greedy AlgorithmPre-computeC[u]: Queries covered by URL u
S[u]
: Spill of URL u w.r.t. R AlgorithmSlide17
Spill Evaluation
Spill queries may be relevant to RDivide spill(a, R) to positive: relevant negative: irrelevantUse query clustering for evaluationExample:a := u
2 u3
Positive spill = {q1}Negative spill = {q5}Slide18
Experimental Evaluation
GoalsCompare output URL bidding variations1-URL, 2-URL, 3-URLe.g, 2-URL: use only URLs, up to 2 URLs in a disjunct1-host, 2-host, 3-host1-mixed, 2-mixedComparison criteriaCompactness – Spill tradeoff
Spill EvaluationSlide19
Experimental Evaluation
SetupDataset (from Yahoo query logs)12,931,117 queries62,666,514 URLs7,185,392 hosts2,251 adsProcessFor each variation (1-URL, 2-URL, …)For different γ values
Generate output expressions for all 2,251 adsSlide20
Experimental Evaluation
Compactness vs SpillSlide21
Experimental Evaluation
Positive and Negative SpillSlide22
Experimental Evaluation
SummaryCompactness-spill trade-offUsing both URLs/hosts outperform other optionsUp to 2 conjuncts in a disjunct is sufficientSpill evaluationOutput expressions can bring additional queriesOther experiments in Combining keyword and output bidding
Output expression are suitable for half of the keywordsUsing only hosts seems to be sufficientSlide23
Conclusions
Output URL bidding can be implemented efficientlyAdvantages over keyword biddingBid CompactnessMore relevant queriesSlide24
THANK YOU!