2. Traditional Advertising
At the beginning: Traditional Ads
Posters, Magazines, Newspapers,
Billboards.
What is being Sold:
Pay-per-Impression: Price depends on
how many people your ad is shown to
(whether or not they look at it)
3. Advertising on the Web
Online Ads:
Banner Ads, Sponsored Search Ads, Pay-per-Sale ads.
Targeting:
Show to particular set of viewers.
Measurement:
Accurate Metrics: Clicks, Tracked Purchases.
What is being Sold:
Pay-per-Click, Pay-per-Action, Pay-per-Impression
4. Banner Ads
Banner ads (1995-2001)
Popular websites charged X$ for
every 1,000 “impressions” of the ad
Called “CPM” rate (Cost per
thousand impressions)
Low click-through rates
Low ROI for advertisers
7. Textual Ads
Web Advertising covers large amount of internet environment.
Internet advertiser spend 17 billion dollars in US on advertising, with 20%
growth rate.
Textual ads covers huge portion of this market.
1. Sponsored Search or paid search advertising
2. Contextual advertising or context match
8. What is AdWords
Advertising system in which advertisers bid on certain keywords in order for their clickable
ads to appear in Google's search results
11. Understanding Adwords Technicalities
Keywords:
Ads are continuously matched to internet users interest based on your keywords
Use keywords match types to your advantage
Broad Match
Phrase Match
Exact Match
Negative Keywords
URL:
Two types of URL are used in an Adwords namely
Display URL
Actual URL
Broad Match
Keywords: buy flowers
Queries:
• buy flowers
• buy red flowers
• flowers buy
• New York buy flowers
Phrase Match
Keywords: “buy flowers”
Queries:
Where can I buy flowers
buy flowers in New Delhi
buy red flowers (extra word)
flowers buy (the words are
reversed)
Exact Match
Keywords: [buy flowers]
Queries:
buy flowers
Buy flowers (Capitalization
doesn’t matter)
Negative Match
Keywords: -cheap
Queries:
buy cheap flowers
cheap flowers in New York
12. Performance based advertising
Introduced by Overture around 2000
Advertisers bid on search keywords
When some one searches for that keywords, the highest bidder’s ad is shown
Advertisers is charged only if the ad is clicked on
Similar model adopted by Google with some changes around 2002
Called adwords
13. Adwords Problem Definition
Input
Advertisers Bid Set for each search query
CTR (Click through Rate) is provided for each pair of query-advertiser.
The budget of each advertiser for time period (it can be month, year
or depends).
Limit on the amount of ads to be illustrated for each search query.
𝑪𝑻𝑹 =
𝑵𝒖𝒎𝒃𝒆𝒓 𝒐𝒇 𝒄𝒍𝒊𝒄𝒌𝒔 𝒐𝒏 𝒂𝒅𝒔
𝑵𝒖𝒎𝒃𝒆𝒓 𝒐𝒇 𝒕𝒊𝒎𝒆𝒔 𝒕𝒉𝒆 𝒂𝒅 𝒊𝒔 𝒔𝒉𝒐𝒘𝒏
14. Adwords Problem Definition
Response for apiece query with advertisers set such that:
The resultant set of ads should not be larger than the provided limit.
Every advertiser has to bid on the query.
For each click on the ads, every advertiser has sufficient budget to pay for it.
𝑹𝒆𝒗𝒆𝒏𝒖𝒆 = 𝑏𝑖𝑑 ∗ 𝐶𝑇𝑅
𝑪𝒐𝒎𝒑𝒆𝒕𝒊𝒕𝒊𝒗𝒆 𝑹𝒂𝒕𝒊𝒐 =
minimum total revenue of algorithm for any sequence
𝑟𝑒𝑣𝑒𝑛𝑢𝑒 𝑜𝑓 𝑜𝑝𝑡𝑖𝑚𝑢𝑚 𝑜𝑓𝑓𝑙𝑖𝑛𝑒 𝑎𝑙𝑔𝑜𝑟𝑖𝑡ℎ𝑚 𝑓𝑜𝑟 𝑡ℎ𝑒 𝑠𝑎𝑚𝑒 𝑠𝑒𝑞𝑢𝑒𝑛𝑐𝑒
The main objective: Maximize the search engine revenue
16. The Adwords Innovation
Advertiser Bid CTR Expected Revenue=Bid*CTR
B $0.75 2% 1.5 Cent
C $0.50 2.5% 1.125 Cent
A $1.00 1% 1 Cent
Instead of sorting advertisers by bid, sort by expected revenue!
• CTR of ad is unknown
• Advertisers have limited budgets and bid on multiple ads (Balance Algo.)
17. Greedy Algorithm
Simplified Environment
There is 1 ad shown for each query
All advertisers have same budget B
All ads are equally likely to be clicked
Value of each ads is the same (=1)
Greedy Algorithm
For a query, select any advertiser who value is 1 for that query.
Competitive Ratio=½.
18. Worst Scenario for Greedy Algorithm
Two Advertiser A and B
A bids on query x, B bids for both query x and y
Both have budget $4
Query Stream: xxxx yyyy
Worst case greedy choice: BBBB _ _ _ _
Optimal: AAAA BBBB
Competitive Ratio =1/2.
This is the worst case!
19. Balance Algorithm
This algorithm proposed by Mehta, Saberi, Vazirani, and Vazirani
For each query, to select the advertiser with the largest unspent budget
Breaks ties arbitrarily
20. Example
Two Advertiser A and B
A bids on query x, B bids for both
query x and y
Both have budget $4
Query Stream: xxxx yyyy
Balance Choice : ABABBB _ _
Optimal: AAAA BBBB
Competitive Ratio =3/4.
For balance with 2 advertisers
(_)
A B
4 4
q(x) A
A B
3 4
q(xx) B
A B
3 3
q(xxx) A
A B
2 3
q(xxxx) B
A B
2 2
q(xxxxy) B (NV: A)
A B
2 1
22. Dataset
Open Advertising dataset
Keywords List
Bidding List: US and UK Market
Bidding dataset1: 177 US Market, 179 UK Market
Bidding dataset2: 244 US Market, 244 UK Market
Bidding dataset3:
Web Page results associated with keywords
26. Reference
Book:
Mining Massive Dataset
By Jure Leskovec, Anand Rajaraman, Jeff Ullman
Papers:
A. Broder, M. Fontoura, V. Josifovski, and L. Riedel, “A semantic approach to
contextual advertising,” Proceedings of the 30th annual international ACM
SIGIR conference on Research and development in information retrieval - SIGIR
’07, 2007.
T.-K. Fan and C.-H. Chang, “Blogger-Centric Contextual Advertising,” Expert
Systems with Applications, vol. 38, no. 3, pp. 1777–1788, Mar. 2011.