References Challenges in Measuring Online Advertising Systems by Saikat Guha Bin Cheng and Paul Francis Serving Ads from localhost for Performance Privacy and Profit by Saikat ID: 1002686
Download Presentation The PPT/PDF document "Lecture 21: Privacy and Online Advertisi..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. Lecture 21: Privacy and Online Advertising
2. ReferencesChallenges in Measuring Online Advertising Systems by Saikat Guha, Bin Cheng, and Paul FrancisServing Ads from localhost for Performance, Privacy, and Profit by Saikat Guha, Alexey Reznichenko, Kevin Tang, Hamed Haddadi, and Paul Francis
3. ProblemOnline advertising funds many web servicesE.g., all the free stuff we get from GoogleAd networks gather much user informationHow do they use the user information?
4. GoalsDetermining how well ad networks target users
5. MethodologyCreating two clients representing two different user typesMeasuring the different ads each client sees
6. ChallengesHow to compare adsHow to collect a representative snapshot of adsQuantifying the differencesAvoiding measurement artifacts
7. Comparing Ads is challengingAds don’t have unique IDsA & B are semantically the same, but with different textA & C are different, but with same display URLs
8. How to define two ads are the same?Easy but illegal approach: comparing destination URLsFP: flagged as equal but notFN: equal but not flaggedDisplay URL has the lowest FNs Use display URL to define ads equality
9. Taking a SnapshotMore ads can be displayed on any single pageHow to determine all Ads that may be fed to a user?Reload the page multiple timesBut too many reloads may lead to ads churn: old ads expire, new ads show up
10. Determining the # of reloadsReloads every 5 secondsRepeated for 200 queriesCurve becomes linear > 10 reloadsAds churnsUse 10 reloads as the threshold
11. Quantifying ChangeMetricsJaccard index: Extended Jaccard index (cosine similarity)
12. Comparing EffectivenessViews: # of page reloads containing the adValue: # of page reloads scaled by the position of the adOverlap: Jaccard index
13. Comparing Effectiveness
14. The winner isWeight: log(views) or log(value)
15. Avoiding artifactsDifferent system parameters may lead to different ads viewBrowsers used different DNS serversBrowsers receive different cookiesHTTP proxy
16. AnalysisConfigure two or more instances to differ by one parameterComparing results forSearch AdsWebsite AdsOnline Social Network Ads
17. Search AdsA, B: control w/o cookiesC, D: w/ cookies enabled. Seeded w/ different personaeGoogle 730 random product-related queries for 5 daysNo obvious behavioral targeting in search ads. Why?Keyword based ads biddingLocation targeting not studied
18. Websites AdsMeasure 15 websites that show Google adsA, B: control in NYC: SF; D: GermanyLocation affects web ads
19. Website AdsA, B: controlC: browse 3 out of 15 websitesD and E: browse random websites and Google search random websitesGoogle does not use browsing behavior to pick ads
20. Online social network adsSet up three or more Facebook profiles A, B: control and identicalC: differs from A by one profile parameter
21.
22.
23. Online social network adsUse all profile parameters to customize adsAge and gender are two primary factorsDiurnal patterns due to ads churnShould it increase or decrease?Education and relationship matter less, except for engaged and non-engaged women
24. Checking Impact of Sexual PreferenceSix profiles with different sexual preferencesTwo males interested in females (male control)Two females interested in males (female control)One male interested in male One female interested in female
25. Ads differ by sexual preferences
26. Other resultsFound neutral ads targeted exclusively to gay menClicking would reveal to the advertiser a user’s sexual preference66 ads shown exclusively to gay men more than 50 times during experiments
27. SummarySearch ads are largely key-word based so farWebsites ads use location but probably not behaviorSocial network ads use all profile attributes to target users
28. Question: how can we design a privacy-preserving online advertising system?
29. GoalsSupport online advertisingA good revenue source to fund online servicesPreserve user privacy
30. PrivAdServing Ads from a localhost clientActors: user, publisher, advertiser, broker, and dealer
31. How it worksAdvertisers upload ads to brokerUser client subscribes to a set of the ads according to the user’s profile to the brokerMessage encrypted with Broker’s public key and contains a symmetric private keyThe Broker sends filtered ads to the user clientAds are encrypted with the symmetric keyDealer anonymizes the client’s message to Broker
32. Ad View/Click ReportingWhen a user clicks an ad, the user client sends a view/click report containing ad ID and publisher ID to the broker via the dealerDealer attaches a unique report ID, removes client identity information, maps the ID to the user identity information
33. Click-fraud defenseBroker provides dealer the record IDs if it suspects click-fraudThe dealer finds the userThe dealer stops relaying ads to user if convincedQuestions not answered: how to detect by broker, and what’s the punishment
34. Defining User PrivacyUnlinkabilityNo single player can link the identity of user with any piece of user’s profileNo single player can link together more than some limited number of pieces of personalization information of a given userThe dealer learns User A clicks on some adThe broker learns someone clicked on ad XNot robust to dealer/broker collusion
35. Scaling PrivAd Ads churn is significant2GB/month of compressed ad data
36. DiscussionWhat challenges does PrivAd may face in a practical deployment?