libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf ·...

118
Sponsored Search

Transcript of libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf ·...

Page 1: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Page 2: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approach

I Keyword-Based AdvertisingI Pay-per-Click

Pay-per-Impression

I Ads not relevant to the pageI Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 3: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approach

I Keyword-Based AdvertisingI Pay-per-Click

Pay-per-Impression

I Ads not relevant to the pageI Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 4: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approach

I Keyword-Based AdvertisingI Pay-per-Click

Pay-per-Impression

I Ads not relevant to the pageI Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 5: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approach

I Keyword-Based AdvertisingI Pay-per-Click

Pay-per-ImpressionI Ads not relevant to the page

I Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 6: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approach

I Keyword-Based AdvertisingI Pay-per-Click

Pay-per-ImpressionI Ads not relevant to the pageI Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 7: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approach

I Keyword-Based AdvertisingI Pay-per-Click

Pay-per-ImpressionI Ads not relevant to the pageI Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 8: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approachI Keyword-Based Advertising

I Pay-per-Click

Pay-per-ImpressionI Ads not relevant to the pageI Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 9: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approachI Keyword-Based AdvertisingI Pay-per-Click

Pay-per-ImpressionI Ads not relevant to the pageI Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 10: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approachI Keyword-Based AdvertisingI Pay-per-Click

Pay-per-ImpressionI Ads not relevant to the pageI Pay even if ad is not useful

Does it work?

I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 11: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approachI Keyword-Based AdvertisingI Pay-per-Click

Pay-per-ImpressionI Ads not relevant to the pageI Pay even if ad is not useful

Does it work?I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60

I The top spot for loan consolidation costs $50

2

Page 12: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Keyword Advertising

The Web Search approachI Keyword-Based AdvertisingI Pay-per-Click

Pay-per-ImpressionI Ads not relevant to the pageI Pay even if ad is not useful

Does it work?I The top spot for calligraphy pens costs $1.50I The top spot for calligaphy pens costs $0.60I The top spot for loan consolidation costs $50

2

Page 13: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How the prices are chosen?

Preliminary definitionsI Clickthrough rates rj of spot j

I Clickthrough rates are knownI Clickthrough rates do not depend on ad quality or relevanceI Clickthrough rates do not depend on ads on other slots

I Revenue per click vi for advertiser iI Advertiser i expected welfare from slot j = vi rj

I This is the maximum amount i is willing to pay for jI Goal: To maximize social welfare

I Best spot to the best advertiser, second best spot to secondbest advertiser, . . .

Problem: we do not know who is the best advertiser

3

Page 14: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How the prices are chosen?

Preliminary definitionsI Clickthrough rates rj of spot j

I Clickthrough rates are knownI Clickthrough rates do not depend on ad quality or relevanceI Clickthrough rates do not depend on ads on other slots

I Revenue per click vi for advertiser iI Advertiser i expected welfare from slot j = vi rj

I This is the maximum amount i is willing to pay for jI Goal: To maximize social welfare

I Best spot to the best advertiser, second best spot to secondbest advertiser, . . .

Problem: we do not know who is the best advertiser

3

Page 15: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How the prices are chosen?

Preliminary definitionsI Clickthrough rates rj of spot j

I Clickthrough rates are knownI Clickthrough rates do not depend on ad quality or relevanceI Clickthrough rates do not depend on ads on other slots

I Revenue per click vi for advertiser i

I Advertiser i expected welfare from slot j = vi rjI This is the maximum amount i is willing to pay for j

I Goal: To maximize social welfareI Best spot to the best advertiser, second best spot to second

best advertiser, . . .

Problem: we do not know who is the best advertiser

3

Page 16: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How the prices are chosen?

Preliminary definitionsI Clickthrough rates rj of spot j

I Clickthrough rates are knownI Clickthrough rates do not depend on ad quality or relevanceI Clickthrough rates do not depend on ads on other slots

I Revenue per click vi for advertiser iI Advertiser i expected welfare from slot j = vi rj

I This is the maximum amount i is willing to pay for j

I Goal: To maximize social welfareI Best spot to the best advertiser, second best spot to second

best advertiser, . . .

Problem: we do not know who is the best advertiser

3

Page 17: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How the prices are chosen?

Preliminary definitionsI Clickthrough rates rj of spot j

I Clickthrough rates are knownI Clickthrough rates do not depend on ad quality or relevanceI Clickthrough rates do not depend on ads on other slots

I Revenue per click vi for advertiser iI Advertiser i expected welfare from slot j = vi rj

I This is the maximum amount i is willing to pay for jI Goal: To maximize social welfare

I Best spot to the best advertiser, second best spot to secondbest advertiser, . . .

Problem: we do not know who is the best advertiser

3

Page 18: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How the prices are chosen?

Preliminary definitionsI Clickthrough rates rj of spot j

I Clickthrough rates are knownI Clickthrough rates do not depend on ad quality or relevanceI Clickthrough rates do not depend on ads on other slots

I Revenue per click vi for advertiser iI Advertiser i expected welfare from slot j = vi rj

I This is the maximum amount i is willing to pay for jI Goal: To maximize social welfare

I Best spot to the best advertiser, second best spot to secondbest advertiser, . . .

Problem: we do not know who is the best advertiser

3

Page 19: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How can we solve this problem?

TruthfulnessI Each advertiser i submit a bid biI It is a dominant strategy for an advertiser i to bid truthfully,

i.e. bi = vi

Can we run second price auctions?I They have been defined for single item auctionsI Now we have multiple slots to sell

4

Page 20: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How can we solve this problem?

TruthfulnessI Each advertiser i submit a bid biI It is a dominant strategy for an advertiser i to bid truthfully,

i.e. bi = vi

Can we run second price auctions?I They have been defined for single item auctionsI Now we have multiple slots to sell

4

Page 21: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

How can we solve this problem?

TruthfulnessI Each advertiser i submit a bid biI It is a dominant strategy for an advertiser i to bid truthfully,

i.e. bi = vi

Can we run second price auctions?I They have been defined for single item auctionsI Now we have multiple slots to sell

4

Page 22: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Second Price Auctions revisitedI The item is allocated to the

highest bidder

I The winner is charged thesecond highest bid

lent tothe harm her presencecauses to the other bidders

I Allocation maximizes thesocial welfare (w.r.t. bids)

I Any agent is charged anamount that is equivalent tothe harm her presencecauses to the other biddersExamples

I n agents with valuation v1 ≥ · · · ≥ vn

I If agent 1 is in the auctionI she wins the itemI remaining player have utility 0

I If agent 1 is not in the auctionI agent 2 wins the item and has utility v2I remaining player have utility 0

I The harm caused by agent 1 amounts to v2

5

Page 23: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Second Price Auctions revisitedI The item is allocated to the

highest bidder

I The winner is charged thesecond highest bid

lent tothe harm her presencecauses to the other bidders

I Allocation maximizes thesocial welfare (w.r.t. bids)

I Any agent is charged anamount that is equivalent tothe harm her presencecauses to the other biddersExamples

I n agents with valuation v1 ≥ · · · ≥ vn

I If agent 1 is in the auctionI she wins the itemI remaining player have utility 0

I If agent 1 is not in the auctionI agent 2 wins the item and has utility v2I remaining player have utility 0

I The harm caused by agent 1 amounts to v2

5

Page 24: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Second Price Auctions revisitedI The item is allocated to the

highest bidderI The winner is charged the

second highest bid

lent tothe harm her presencecauses to the other bidders

I Allocation maximizes thesocial welfare (w.r.t. bids)

I Any agent is charged anamount that is equivalent tothe harm her presencecauses to the other biddersExamples

I n agents with valuation v1 ≥ · · · ≥ vn

I If agent 1 is in the auctionI she wins the itemI remaining player have utility 0

I If agent 1 is not in the auctionI agent 2 wins the item and has utility v2I remaining player have utility 0

I The harm caused by agent 1 amounts to v2

5

Page 25: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Second Price Auctions revisitedI The item is allocated to the

highest bidderI The winner is charged the

second highest bid

lent tothe harm her presencecauses to the other bidders

I Allocation maximizes thesocial welfare (w.r.t. bids)

I Any agent is charged anamount that is equivalent tothe harm her presencecauses to the other bidders

ExamplesI n agents with valuation v1 ≥ · · · ≥ vn

I If agent 1 is in the auctionI she wins the itemI remaining player have utility 0

I If agent 1 is not in the auctionI agent 2 wins the item and has utility v2I remaining player have utility 0

I The harm caused by agent 1 amounts to v2

5

Page 26: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Second Price Auctions revisitedI The item is allocated to the

highest bidderI The winner is charged the

second highest bid

lent tothe harm her presencecauses to the other bidders

I Allocation maximizes thesocial welfare (w.r.t. bids)

I Any agent is charged anamount that is equivalent tothe harm her presencecauses to the other biddersExamples

I n agents with valuation v1 ≥ · · · ≥ vn

I If agent 1 is in the auctionI she wins the itemI remaining player have utility 0

I If agent 1 is not in the auctionI agent 2 wins the item and has utility v2I remaining player have utility 0

I The harm caused by agent 1 amounts to v2

5

Page 27: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Second Price Auctions revisitedI The item is allocated to the

highest bidderI The winner is charged the

second highest bid

lent tothe harm her presencecauses to the other bidders

I Allocation maximizes thesocial welfare (w.r.t. bids)

I Any agent is charged anamount that is equivalent tothe harm her presencecauses to the other biddersExamples

I n agents with valuation v1 ≥ · · · ≥ vnI If agent 1 is in the auction

I she wins the itemI remaining player have utility 0

I If agent 1 is not in the auctionI agent 2 wins the item and has utility v2I remaining player have utility 0

I The harm caused by agent 1 amounts to v2

5

Page 28: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Second Price Auctions revisitedI The item is allocated to the

highest bidderI The winner is charged the

second highest bid

lent tothe harm her presencecauses to the other bidders

I Allocation maximizes thesocial welfare (w.r.t. bids)

I Any agent is charged anamount that is equivalent tothe harm her presencecauses to the other biddersExamples

I n agents with valuation v1 ≥ · · · ≥ vnI If agent 1 is in the auction

I she wins the itemI remaining player have utility 0

I If agent 1 is not in the auctionI agent 2 wins the item and has utility v2I remaining player have utility 0

I The harm caused by agent 1 amounts to v2

5

Page 29: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Second Price Auctions revisitedI The item is allocated to the

highest bidderI The winner is charged the

second highest bid

lent tothe harm her presencecauses to the other bidders

I Allocation maximizes thesocial welfare (w.r.t. bids)

I Any agent is charged anamount that is equivalent tothe harm her presencecauses to the other biddersExamples

I n agents with valuation v1 ≥ · · · ≥ vnI If agent 1 is in the auction

I she wins the itemI remaining player have utility 0

I If agent 1 is not in the auctionI agent 2 wins the item and has utility v2I remaining player have utility 0

I The harm caused by agent 1 amounts to v2

5

Page 30: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions

Vickrey-Clarke-Groves PrincipleI Agents submit bidsI Allocation maximizes the social welfareI Prices are the harm to other bidders

VCG PricesI SW (A) = maximum social welfare with all bidders and slotsI SW (A−i

−j) = maximum social welfare without j and its slot iI SW (A−j) = maximum social welfare without j (but all slots)

pij = SW (A−j) − SW (A−i−j)

6

Page 31: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions

Vickrey-Clarke-Groves PrincipleI Agents submit bidsI Allocation maximizes the social welfareI Prices are the harm to other bidders

VCG PricesI SW (A) = maximum social welfare with all bidders and slotsI SW (A−i

−j) = maximum social welfare without j and its slot iI SW (A−j) = maximum social welfare without j (but all slots)

pij = SW (A−j) − SW (A−i−j)

6

Page 32: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions

Vickrey-Clarke-Groves PrincipleI Agents submit bidsI Allocation maximizes the social welfareI Prices are the harm to other bidders

VCG PricesI SW (A) = maximum social welfare with all bidders and slotsI SW (A−i

−j) = maximum social welfare without j and its slot iI SW (A−j) = maximum social welfare without j (but all slots)

pij = SW (A−j) − SW (A−i−j)

6

Page 33: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 34: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)

I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 35: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12

I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 36: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25

I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 37: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13

I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 38: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32

I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 39: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35

I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 40: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3

I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 41: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40

I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 42: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40

I The third advertiser must pay 40 − 40 = 0

7

Page 43: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions - ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI Optimal assignment (welfare = 42)I Welfare without first advertiser and first slot: 12I Welfare without first advertiser but with first slot: 25I The first advertiser must pay 25 − 12 = 13I Welfare without second advertiser and second slot: 32I Welfare without the second advertiser but with second slot: 35I The second advertiser must pay 35 − 32 = 3I Welfare without third advertiser and third slot: 40I Welfare without the third advertiser but with third slot: 40I The third advertiser must pay 40 − 40 = 0

7

Page 44: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Truthfulness

I If the lie does not affect the assigned slot, payoff unchangedI By lying, i takes item k in place of item j

vij − pij?≥ vih − pih

vij −[SW (A−i ) − SW (A−j

−i )] ?

≥ vih −[SW (A−i ) − SW (A−h

−i )]

vij + SW (A−j−i )

?≥ vih + SW (A−h

−i )

SW (A)?≥ vih + SW (A−h

−i )

YES

8

Page 45: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Truthfulness

I If the lie does not affect the assigned slot, payoff unchanged

I By lying, i takes item k in place of item j

vij − pij?≥ vih − pih

vij −[SW (A−i ) − SW (A−j

−i )] ?

≥ vih −[SW (A−i ) − SW (A−h

−i )]

vij + SW (A−j−i )

?≥ vih + SW (A−h

−i )

SW (A)?≥ vih + SW (A−h

−i )

YES

8

Page 46: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Truthfulness

I If the lie does not affect the assigned slot, payoff unchangedI By lying, i takes item k in place of item j

vij − pij?≥ vih − pih

vij −[SW (A−i ) − SW (A−j

−i )] ?

≥ vih −[SW (A−i ) − SW (A−h

−i )]

vij + SW (A−j−i )

?≥ vih + SW (A−h

−i )

SW (A)?≥ vih + SW (A−h

−i )

YES

8

Page 47: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Truthfulness

I If the lie does not affect the assigned slot, payoff unchangedI By lying, i takes item k in place of item j

vij − pij?≥ vih − pih

vij −[SW (A−i ) − SW (A−j

−i )] ?

≥ vih −[SW (A−i ) − SW (A−h

−i )]

vij + SW (A−j−i )

?≥ vih + SW (A−h

−i )

SW (A)?≥ vih + SW (A−h

−i )

YES

8

Page 48: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Truthfulness

I If the lie does not affect the assigned slot, payoff unchangedI By lying, i takes item k in place of item j

vij − pij?≥ vih − pih

vij −[SW (A−i ) − SW (A−j

−i )] ?

≥ vih −[SW (A−i ) − SW (A−h

−i )]

vij + SW (A−j−i )

?≥ vih + SW (A−h

−i )

SW (A)?≥ vih + SW (A−h

−i )

YES

8

Page 49: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Truthfulness

I If the lie does not affect the assigned slot, payoff unchangedI By lying, i takes item k in place of item j

vij − pij?≥ vih − pih

vij −[SW (A−i ) − SW (A−j

−i )] ?

≥ vih −[SW (A−i ) − SW (A−h

−i )]

vij + SW (A−j−i )

?≥ vih + SW (A−h

−i )

SW (A)?≥ vih + SW (A−h

−i )

YES

8

Page 50: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Truthfulness

I If the lie does not affect the assigned slot, payoff unchangedI By lying, i takes item k in place of item j

vij − pij?≥ vih − pih

vij −[SW (A−i ) − SW (A−j

−i )] ?

≥ vih −[SW (A−i ) − SW (A−h

−i )]

vij + SW (A−j−i )

?≥ vih + SW (A−h

−i )

SW (A)?≥ vih + SW (A−h

−i )

YES

8

Page 51: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Truthfulness

I If the lie does not affect the assigned slot, payoff unchangedI By lying, i takes item k in place of item j

vij − pij?≥ vih − pih

vij −[SW (A−i ) − SW (A−j

−i )] ?

≥ vih −[SW (A−i ) − SW (A−h

−i )]

vij + SW (A−j−i )

?≥ vih + SW (A−h

−i )

SW (A)?≥ vih + SW (A−h

−i ) YES

8

Page 52: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3 x30, 15, 6 3

y2 y20, 10, 4 2

z1 z10, 5, 2 1

1 101 10 13

2 52 5 3

3 23 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 53: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3 x30, 15, 6 3

y2 y20, 10, 4 2

z1 z10, 5, 2 1

1 101 10 13

2 52 5 3

3 23 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 54: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3 x30, 15, 6 3

y2 y20, 10, 4 2

z1 z10, 5, 2 1

1 101 10 13

2 52 5 3

3 23 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 55: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Marketsx3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 56: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 57: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 58: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 59: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 60: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 61: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price Auctions

I Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 62: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price AuctionsI Market-clearing prices given by generalized ascending auctions

I Ascending Auctions are equivalent to Second-Price Auctions

9

Page 63: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VGC Auctions: Market Clearing PricesVGC pricesPersonalized Prices

Market Clearing pricesPosted Prices

Let us model sponsored search as Matching Markets

x3

x30, 15, 6 3

y2

y20, 10, 4 2

z1

z10, 5, 2 1

1 10

1 10 13

2 5

2 5 3

3 2

3 2 0

VCG prices vs. Market Clearing PricesVCG prices are market clearing prices of minimum total sum

ObservationsI VGC Auctions are a generalization of Second-Price AuctionsI Market-clearing prices given by generalized ascending auctionsI Ascending Auctions are equivalent to Second-Price Auctions

9

Page 64: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Advantages and Drawbacks

AdvantagesI TruthfulnessI Market Clearing PriceI The approach extend to other types of auctions

DrawbacksI Payment rule is hard to describeI It may be expensive to compute paymentsI It maximizes social welfare, but what about revenue?

10

Page 65: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Advantages and Drawbacks

AdvantagesI TruthfulnessI Market Clearing PriceI The approach extend to other types of auctions

DrawbacksI Payment rule is hard to describe

I It may be expensive to compute paymentsI It maximizes social welfare, but what about revenue?

10

Page 66: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Advantages and Drawbacks

AdvantagesI TruthfulnessI Market Clearing PriceI The approach extend to other types of auctions

DrawbacksI Payment rule is hard to describeI It may be expensive to compute payments

I It maximizes social welfare, but what about revenue?

10

Page 67: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

VCG Auctions: Advantages and Drawbacks

AdvantagesI TruthfulnessI Market Clearing PriceI The approach extend to other types of auctions

DrawbacksI Payment rule is hard to describeI It may be expensive to compute paymentsI It maximizes social welfare, but what about revenue?

10

Page 68: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Prices are the harm to other

bidders

GSP Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Price is the valuation of the

next slot winnerExample

I 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 69: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bids

I Allocation maximizes thesocial welfare

I Prices are the harm to otherbidders

GSP Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Price is the valuation of the

next slot winnerExample

I 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 70: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bids

I Allocation maximizes thesocial welfare

I Prices are the harm to otherbidders

GSP AuctionI Agents submit bids

I Allocation maximizes thesocial welfare

I Price is the valuation of thenext slot winner

ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 71: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfare

I Prices are the harm to otherbidders

GSP AuctionI Agents submit bids

I Allocation maximizes thesocial welfare

I Price is the valuation of thenext slot winner

ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 72: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfare

I Prices are the harm to otherbidders

GSP AuctionI Agents submit bidsI Allocation maximizes the

social welfare

I Price is the valuation of thenext slot winner

ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 73: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Prices are the harm to other

bidders

GSP AuctionI Agents submit bidsI Allocation maximizes the

social welfare

I Price is the valuation of thenext slot winner

ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 74: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Prices are the harm to other

bidders

GSP AuctionI Agents submit bidsI Allocation maximizes the

social welfareI Price is the valuation of the

next slot winner

ExampleI 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 75: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Prices are the harm to other

bidders

GSP AuctionI Agents submit bidsI Allocation maximizes the

social welfareI Price is the valuation of the

next slot winnerExample

I 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuations

I The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 76: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Prices are the harm to other

bidders

GSP AuctionI Agents submit bidsI Allocation maximizes the

social welfareI Price is the valuation of the

next slot winnerExample

I 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI The first advertiser pays 2 × 10 = 20 for the first slot

I The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 77: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Prices are the harm to other

bidders

GSP AuctionI Agents submit bidsI Allocation maximizes the

social welfareI Price is the valuation of the

next slot winnerExample

I 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slot

I The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 78: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Generalized Second Price AuctionVCG Auction

I Agents submit bidsI Allocation maximizes the

social welfareI Prices are the harm to other

bidders

GSP AuctionI Agents submit bidsI Allocation maximizes the

social welfareI Price is the valuation of the

next slot winnerExample

I 3 slots with Clickthrough Rates 10, 5 and 2I 3 advertisers with Revenue per Click 3, 2 and 1I Assume bids = valuationsI The first advertiser pays 2 × 10 = 20 for the first slotI The second advertiser pays 1 × 5 = 5 for the second slotI The third advertiser pays 0 × 2 = 0 for the third slot

11

Page 79: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 80: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibrium

I 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 81: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1

I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 82: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10

I If x bids 5, then she has payoff 28 − 4 = 24I There may be multiple equilibria

I Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 83: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 84: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibrium

I Bids bx = 3, by = 5, bz = 1 are in equilibriumI There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 85: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 86: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfare

I There are equilibria with revenue larger than the VCG revenueI (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 87: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48

I Truthful equilibrium gives revenue 44I There are equilibria with revenue lower than the VCG revenue

I (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 88: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 89: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenue

I (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 90: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP Auctions: Drawbacks

I Truthtelling is not an equilibriumI 2 slots: r1 = 10, r2 = 4; 3 bidders: vx = 7, vy = 6, vz = 1I Truthful bidder x has payoff 70 − 60 = 10I If x bids 5, then she has payoff 28 − 4 = 24

I There may be multiple equilibriaI Bids bx = 5, by = 4, bz = 2 are in equilibriumI Bids bx = 3, by = 5, bz = 1 are in equilibrium

I There are equilibria that do not maximize the social welfareI There are equilibria with revenue larger than the VCG revenue

I (bx = 5, by = 4, bz = 2) gives revenue 48I Truthful equilibrium gives revenue 44

I There are equilibria with revenue lower than the VCG revenueI (bx = 3, by = 5, bz = 1) gives revenue 34

12

Page 91: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7 x70, 28, 0 7

y6 y60, 24 6 y60, 24, 0 6

z1 z10, 4 1 z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0 xb∗1 > 4 70, 28, 0

y60, 24, 0 yb∗2 = 4 60, 24, 0

z10, 4, 0 zb∗3 = 1 10, 4, 0

1 401 40 p∗1 = 4

2 42 4 p∗2 = 1

3 03 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 92: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching Market

I Compute bids from Market Clearing Pricesx7 x70, 28 7 x70, 28, 0 7

y6 y60, 24 6 y60, 24, 0 6

z1 z10, 4 1 z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0 xb∗1 > 4 70, 28, 0

y60, 24, 0 yb∗2 = 4 60, 24, 0

z10, 4, 0 zb∗3 = 1 10, 4, 0

1 401 40 p∗1 = 4

2 42 4 p∗2 = 1

3 03 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 93: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching Market

I Compute bids from Market Clearing Prices

x7

x70, 28 7 x70, 28, 0 7

y6

y60, 24 6 y60, 24, 0 6

z1

z10, 4 1 z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0 xb∗1 > 4 70, 28, 0

y60, 24, 0 yb∗2 = 4 60, 24, 0

z10, 4, 0 zb∗3 = 1 10, 4, 0

1 401 40 p∗1 = 4

2 42 4 p∗2 = 1

3 03 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 94: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching Market

I Compute bids from Market Clearing Pricesx7

x70, 28 7

x70, 28, 0 7

y6

y60, 24 6

y60, 24, 0 6

z1

z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0 xb∗1 > 4 70, 28, 0

y60, 24, 0 yb∗2 = 4 60, 24, 0

z10, 4, 0 zb∗3 = 1 10, 4, 0

1 401 40 p∗1 = 4

2 42 4 p∗2 = 1

3 03 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 95: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching Market

I Compute bids from Market Clearing Pricesx7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0 xb∗1 > 4 70, 28, 0

y60, 24, 0 yb∗2 = 4 60, 24, 0

z10, 4, 0 zb∗3 = 1 10, 4, 0

1 401 40 p∗1 = 4

2 42 4 p∗2 = 1

3 03 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 96: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0 xb∗1 > 4 70, 28, 0

y60, 24, 0 yb∗2 = 4 60, 24, 0

z10, 4, 0 zb∗3 = 1 10, 4, 0

1 401 40 p∗1 = 4

2 42 4 p∗2 = 1

3 03 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 97: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 98: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 99: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 100: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 101: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 102: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 103: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 104: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 105: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibrium

I Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NO

I Assigned: There are equilibria with good properties

13

Page 106: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibriumI Lower bid and same slot

NO

I Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NOI Assigned: There are equilibria with good properties

13

Page 107: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibriumI Lower bid and same slot NOI Lower bid and lower slot

NO

I Higher bid and same slot

NO

I Higher bid and higher slot

NOI Assigned: There are equilibria with good properties

13

Page 108: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibriumI Lower bid and same slot NOI Lower bid and lower slot NOI Higher bid and same slot

NO

I Higher bid and higher slot

NOI Assigned: There are equilibria with good properties

13

Page 109: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibriumI Lower bid and same slot NOI Lower bid and lower slot NOI Higher bid and same slot NOI Higher bid and higher slot

NOI Assigned: There are equilibria with good properties

13

Page 110: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibriumI Lower bid and same slot NOI Lower bid and lower slot NOI Higher bid and same slot NOI Higher bid and higher slot NO

I Assigned: There are equilibria with good properties

13

Page 111: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

GSP AuctionDoes an equilibrium exists?

I Sponsored Search as a Matching MarketI Compute bids from Market Clearing Prices

x7 x70, 28 7

x70, 28, 0 7

y6 y60, 24 6

y60, 24, 0 6

z1 z10, 4 1

z10, 4, 0 1

1 10

2 4

3 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

x70, 28, 0

xb∗1 > 4 70, 28, 0

y60, 24, 0

yb∗2 = 4 60, 24, 0

z10, 4, 0

zb∗3 = 1 10, 4, 0

1 40

1 40 p∗1 = 4

2 4

2 4 p∗2 = 1

3 0

3 0 p∗3 = 0

p∗1 ≥ p∗

2 ≥ . . . ≥ p∗n

I These bids are in equilibriumI Lower bid and same slot NOI Lower bid and lower slot NOI Higher bid and same slot NOI Higher bid and higher slot NO

I Assigned: There are equilibria with good properties

13

Page 112: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Ad Quality

I We assumed that the Clickthrough Rate depends only on slot

I High quality ads attract more clicks (and larger revenue)I Embedding ad quality in the Sponsored Search model

I Each ad has a quality factor qjI Clickthrough Ratio of ad j in slot i : qj riI Advertiser j valuation for slot i : vij = vjqj ri

I This change does not affect the properties of auctions

14

Page 113: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Ad Quality

I We assumed that the Clickthrough Rate depends only on slotI High quality ads attract more clicks (and larger revenue)

I Embedding ad quality in the Sponsored Search modelI Each ad has a quality factor qjI Clickthrough Ratio of ad j in slot i : qj riI Advertiser j valuation for slot i : vij = vjqj ri

I This change does not affect the properties of auctions

14

Page 114: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Ad Quality

I We assumed that the Clickthrough Rate depends only on slotI High quality ads attract more clicks (and larger revenue)

I Embedding ad quality in the Sponsored Search model

I Each ad has a quality factor qjI Clickthrough Ratio of ad j in slot i : qj riI Advertiser j valuation for slot i : vij = vjqj ri

I This change does not affect the properties of auctions

14

Page 115: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Ad Quality

I We assumed that the Clickthrough Rate depends only on slotI High quality ads attract more clicks (and larger revenue)

I Embedding ad quality in the Sponsored Search modelI Each ad has a quality factor qj

I Clickthrough Ratio of ad j in slot i : qj riI Advertiser j valuation for slot i : vij = vjqj ri

I This change does not affect the properties of auctions

14

Page 116: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Ad Quality

I We assumed that the Clickthrough Rate depends only on slotI High quality ads attract more clicks (and larger revenue)

I Embedding ad quality in the Sponsored Search modelI Each ad has a quality factor qjI Clickthrough Ratio of ad j in slot i : qj ri

I Advertiser j valuation for slot i : vij = vjqj ri

I This change does not affect the properties of auctions

14

Page 117: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Ad Quality

I We assumed that the Clickthrough Rate depends only on slotI High quality ads attract more clicks (and larger revenue)

I Embedding ad quality in the Sponsored Search modelI Each ad has a quality factor qjI Clickthrough Ratio of ad j in slot i : qj riI Advertiser j valuation for slot i : vij = vjqj ri

I This change does not affect the properties of auctions

14

Page 118: libeccio.di.unisa.itlibeccio.di.unisa.it/SocialNetworks_2015/slide/sponsored_search.pdf · Sponsored Search How the prices are chosen? Preliminary definitions I Clickthrough rates

Sponsored Search

Ad Quality

I We assumed that the Clickthrough Rate depends only on slotI High quality ads attract more clicks (and larger revenue)

I Embedding ad quality in the Sponsored Search modelI Each ad has a quality factor qjI Clickthrough Ratio of ad j in slot i : qj riI Advertiser j valuation for slot i : vij = vjqj ri

I This change does not affect the properties of auctions

14