sponsored search markets - cs.bgu.ac.ilatcs131/wiki.files/lihi... · פרסום באינטרנט...
TRANSCRIPT
Sponsored Search Markets
Advertising Tied to Search Behavior
• Early Web advertising was sold on the basis of
“impressions,” by analogy with the print ads one sees in newspapers or magazines.
A company would negotiate a rate with an advertiser, agreeing on a price for showing its ad a fixed number of times.
The problem..
- The ad that is showing a users isn’t always tied in some intrinsic way to their behavior.
- The advertiser paying to display ads to the full iInternet using population seems like a very iiiinefficient way to find customers.
For example
A very small retailer who’s trying to run a business that sells PILOT pens.
- Preferred work out an agreement with a search engine that said,
“Show my ad to users who enter the query ‘PILOT pens.’ ”
- Search engine queries are a potent way to get users to express their intent
An ad that is based on the query is catching a user at precisely this receptive moment.
• Today - keyword-based advertising.
Advertising revenue 2012
In 2012 the internet is still the most growing category
Google v.s new York times
• While in 2002 Google Revenue from ads NYT. Was 1/6 of
• Today Google Revenue are
NYT times20 then more
New York Times
Who are the players and what are their interest?
8
to maximize
revenue
Get people who will acquire their
product and pay the minimum price
Find best product information
users
advertisers
“keyword-based” pricing Paying per click
• Cost per click model.
• You only pay when a user actually click on the ad.
• Clicking on an ad represent a stronger indication of intent then of simply issuing a query.
Some times can reach to more then 50$ per click
“mortgage refinancing,”
Keuka lake
up to 50$ Up to 1.5$
How to set the price per click for different queries?
One way:
The search engine posts prices (like prices of items in a store).
The problem:
So many possible keywords and combinations
of keywords, make it impossible to decide prices for all these queries.
the solution auction
The seller goal The problem
to maximize
revenue Advertisers want
To pay the less they can and willing to lie
Building model which encouraging
truthful bidding
,VCG GSP
But still there is a conflict between !the buyer and the seller
Slots – the set of available locations that the
search engine has for selling ads on a given query
Slot a Slot
b
Slot c
Slot d
…
…
Click through rate — this is the number of clicks per hour that an ad placed in that slot will receive.
Assuming:
1. The advertisers know the click through rates.
2. Click through rate depends only on the slot itself and not on the ad that is placed there.
3. The click through rate of a slot also doesn’t depend on the ads that are in other slots.
Click Through rate
9%
4%
2%
0.5%
0.2%
0.08%
Revenue per click - the expected
amount of revenue the advertiser receives per user who clicks on the ad.
Advertising as a Matching Market
Constructing a Matching Market
Matching Market:
• The participants in a matching market consist of a set of buyers and a set of sellers.
• Each buyer j has a valuation for the item offered by each seller i.
Constructing a Matching Market
The goal
• Match up buyers with sellers.
• No buyer purchases two different items (slots).
• The same item (slot) isn’t sold to two different buyers.
we define: makteIn this matching
• ri - clickthrough rate of slot i
• vj - the revenue per click of advertiser j
rivj - The benefit that advertiser j receives from llllllbeing shown in slot i.
- The valuation for slot i of advertiser j
seller buyer
Click through rates * revenues per click = valuations
a
c
b rivj
ri vj
unequal numbers of slots and advertisers
• If there are more advertisers than slots, we simply create additional “fictitious” slots of clickthrough rate 0
• If there are more slots than advertisers, we simply create additional “fictitious” advertisers with revenues 0
Obtaining Market Clearing Prices • The price are market clearing if we get a graph has a
perfect matching: in this case, we can assign distinct items to all the buyers in such a way that each buyer gets an item that maximizes her payoff
• Each seller i announces a price pi for his item.
• Each buyer j evaluates her payoff for choosing a
particular seller i (= vij – Pi)
• We then build a preferred seller graph by linking each buyer to the seller or sellers from which she gets the highest payoff.
Click through rates * revenues per click = valuations
buyer X bought
slot a (he's max payoff):
1. X cannot buy slot b,c. 2. Y,Z cannot buy slot a.
This construction of prices can only be carried out by a search engine if it actually knows the valuations of the advertisers.
How to set prices in a setting where the search engine doesn’t know these valuations?
example - second - price auctions
• Each bidder writes his bid in a sealed envelope.
• The seller:
– Collects bids
– Open envelopes.
• Winner: bidder with the highest bid.
Payment: winner pays the 2nd highest bid.
Note: bidders do not see the bids of the other bidders.
$2 $3 $8
at $5
$5
Second price single item auction principles:
• Encourages truthful bidding.
• Produces an allocation that maximizes social welfare - the bidder who values the item the most gets it.
• The winner is charged an price that equal to the “harm” he causes the other bidders by receiving the item.
The presence of bidder A causes the “harm” of 5$ to bidder C. (Bidder C lost an item worthy of 5$)
Therefore, bidder A must pay 5$ as the price.
$2 $3 $8
at $5
$5
A B C D
How can we define a price setting procedure for matching market with this principles?
VCG – Vickerey Clarke Groves
Procedure: • Buyers announce valuations for the items.
• Bidders do not see the bids of the other bidders.
• The assignment - the max valuation for a slot wins the slot. This is the socially optimal assignment of items to buyers — that is, a perfect matching that maximizes the total valuation of each buyer for what they get.
• Compute price with the VCG principles.
What prices does the VCG principle dictate for each buyer?
Buyer j should pay for seller i’s item (slot) the harm she causes to the remaining buyers through her acquisition of this item.
rivj
in the optimal matching without buyer x present, buyer y gets item a and buyer z gets item b. This improves the respective valuations of y and z for their assigned items by 20 − 10 = 10 and 5 − 2 = 3 respectively. The total harm caused by x is therefore 10 + 3 = 13, and so this is the price that x should pay.
20 − 10 = 10
5− 2 = 3
10+ 3= 13
price in the VCG model