1 / 19

Towards a Generic Bidding Standard for Online Advertising

Sihem Amer-Yahia Sebastien Lahaie David Pennock Yahoo! Research. Towards a Generic Bidding Standard for Online Advertising. Advertising Then and Now. Then: Think real estate Phone calls Manual negotiation “Half doesn’t work”.

Download Presentation

Towards a Generic Bidding Standard for Online Advertising

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Sihem Amer-Yahia Sebastien Lahaie David Pennock Yahoo! Research Towards a Generic Bidding Standard for Online Advertising

  2. Advertising Then and Now • Then: Think real estatePhone callsManual negotiation“Half doesn’t work” • Now: Think Wall StreetComputer learns what ad is bestComputer mediates ad sales: AuctionComputer measures which ads workAdvertisers buy contextual events: User i views/clicks/converts on page j at time t automation automation automation

  3. Today and Tomorrow • Today:Multiple competing “black box” networksProprietary data and interfacesSilo-ed ad types (display, search, contextual, affiliate), largely historical artifacts • Tomorrow (hopefully):Open standard bidding languageFor advertisers and publishersPortable among ad networksUnified ad types

  4. The Pitch • An open standard for advertisers (& publishers) to describe their campaign goals that • Is expressive -- lets advertisers state different value for different (bundles of) contextual events • Is extensible -- doomed to fail otherwise • Is efficient -- is amenable to tractable allocation and pricing algorithms

  5. Why? • Advertisers benefit most • Portability, choice • Flexibility, targeting • Small publishers/networks & ad agencies: Reduced startup/switching costs • Technology providers: New markets for analytics, campaign management tools • Users: Better ads, maybe • Academics: Opens up black boxes

  6. Why? Because. • Open standards are flourishing despite non-obvious economics • Open source, wikinomics • APIs, widgets, mashups • Microformats • End of DRM • DomainKeys • OpenX.com • Openness one of Yahoo!’s three “Big Bets”, and the only economically non-obvious one 6

  7. Why Not? • Largest networks & ad agencies may lose their “lock in”, have little incentive to join • Sounds good in theory -- is a common standard even possible in practice? Microformats hard enough -- an ad language is massively more complex • Online ad space evolving too fast -- not ready for standards yet • Some features of current networks (e.g., real-time bidding based on “pixeled” users) inevitably will not be part of the standard 7

  8. Language design goals • Expressive • DG1: Advertisers should be able to specify the type of campaign they want to instigate • DG2: Advertisers should be able to specify their willingness to pay for individual contextual events • DG3: Advertisers should be able to specify volume constraints over bundles of contextual events • DG4: Advertisers should be able to specify budget constraints for their campaigns • DG5: Advertisers should be able to specify contingency upon which payment will be made • DG10: Publishers should be able to formally state the semantics used to interpret instances of the language • Extensible • DG6: The language should be extensible with respect to new targeting dimensions • DG7: The language should be extensible with respect to new bidding mechanisms • Efficient • DG8: The language should allow for static checking of unrecognized or forbidden targeting dimensions, forbidden bids, and forbidden volume constraints • DG9: The language should allow for efficient allocation and pricing algorithms

  9. Ad Serving Engine Advertisers Publishers Ad Networks Ad campaigns in XML Pricing, allocation, spend management algorithms common data structure compiler Ad processing platform Get inventory Update budget Update impressions Standard Data Access Layer Ad inventory database

  10. Language definition and semantics • A campaign is associated to an advertiser and is described by: • adv is an advertiser id type is the campaign type (SS,DA,CA) id is the campaign identifier Budget is the advertiser's budget expr is an expression identifying an event (signal) • Each expr is of the form:

  11. Tree structure Forest (additive) Google, Microsoft, Yahoo! offer this Nesting

  12. Nesting In general, nodes further annotated with volume constraints

  13. Language syntax: XML • Flexible syntax which allows optional elements (needed for dimensions and bids) • Tree structure facilitates nesting contextual events (signals) and helps to target more specific population segments • Wide availability of XML tools (e.g., parsers and query engines) facilitates data processing

  14. Display Advertising <campaign type=“DA" adv="156" id="adv156-1" budget="$1000"> <description>sale to young adults</description> <creative>Buy one get one free</creative> <expr> <dimension name="age" op= "=" value="[25-34]"/> <bids> <cpm val="0.4" min="0" max="500"/> </bids> </expr> </campaign> Other dimensions: <dimension name="location“ op="=“ values="{CA,FL,WI,IL,TX}"/> <dimension name="time" op="=" value="{am,pm}"/> <dimension name="day" op="=" value="[Fri-Sun]"/> <dimension name="position" op="=" value="south"/> <creative>http://www.nike-camp.com/newSneakers.jpg</>

  15. Display Advertising <campaign type=“DA" adv="156" id="adv156-8" budget="$9000"> <description>sale to young adults</> <creative>Buy one get one free</> <expr> <dimension name="age" op= "=" values="[25-35]"/> <bids> <cpm val="0.3" min="20K" max="100K"/> <cpc val="0.4" min="200"/> <cpa val="0.6"/> </bids> <expr> <dimension name="location" op="=" values="CA"/> <bids> <cpm val="0.3" max="50K"/> </bids> </expr> <expr> <dimension name="location“ op="=“ values="{NY,NJ,PA}"/> <bids> <cpc val="0.4" min="50"/> </bids> </expr> </expr> </campaign>

  16. Sponsored Search <campaign type="SS" adv="5438" id="adv5438-78“ budget="$1000"> <description>Cheap Flights to San Francisco</> <creative>Save Up to 65% On Flights to CA Plus $10 Instant Discount</> <keywords>"flight promotion“ "New York flight“ "San Francisco flight“ "direct flight"</> <expr> <dimension name="location" op="=" val="east-coast"/> <bids> <cpc val="0.2"/> <cpa val="0.5"/> </bids> </expr> </campaign>

  17. Contextual Advertising <campaign type="CA" adv="156" id="adv156-1 "budget="$1500"> <description>male sneaker sale</> <creative>Buy one get one free</> <keywords>"promotion“ "sneakers“ "sports“ "running shoes“ "basketball"</> <expr> <dimension name="gender" op="=" values="male"/> <bids> <cpc val="0.4"/> <cpa val="0.6" max="300"/> </bids> <expr> <dimension name="position" op="=" values="north"> <bids> <cpm val="0.2"/> </bids> </expr> </expr> </campaign>

  18. Efficiency • Static checking for logical errors • Allocation depends on network interpretation: How to convert values on individual contextual events to values on bundles? • Greedy online allocation(sponsored search, ad exchange model) • Volume constraints only => PolytimeSee: “An Expressive Auction Design for Online Display Advertising”, Lahaie et al., AAAI-2008 • Budget constraints only => PolytimeSee Vazirani

  19. Agenda • Implement the language compiler • Design the API • Implement pricing and allocation algorithm using the data access API

More Related