Presentations text content in ADNOSTIC Privacy Preserving targeted advertising.
Slide1
ADNOSTIC
Privacy Preserving targeted advertising.
Presented By:Humam Nameer
1
Slide2CASE STUDY
Just over three years old,
Clash of Clans
rakes in more than $
5M
each day and consistently charts in the top three grossing
apps. It’s the app that new game developers look to for inspiration.Clash did begin investing quite heavily in paid media (some speculate as much as $1M per day).
Adnostic: Privacy Preserving Targeted Audience
2
Slide3Adnostic
: Privacy Preserving Targeted Audience
Table of Contents
Brief Overview of online behavioral advertising
Problem Statement
Goal of
ResearchAdnostic ArchitectureImplementation
EvaluationIncentives Offered by AdnosticPossible ObjectionsConslusion3
Slide4Adnostic
: Privacy Preserving Targeted Audience
Chapter 1
Brief Overview of online behavioral advertising
4
Slide5Adnostic
: Privacy Preserving Targeted Audience
Brief Overview of online behavioral advertising
5
Slide6Adnostic
: Privacy Preserving Targeted Audience
Behavioral Targeting
Ability to serve relevant content / Products / advertisement to user based on their past behavior.
Users are categorized in one or more segments and content from the related segment is server.
Brief Overview of online behavioral advertising
6
Slide7Adnostic
: Privacy Preserving Targeted Audience
What determines
Behaviour
Content Viewed
Entry page & Path Taken
Frequency
Search Keywords
Referring Keywords
Past purchases/ sign ups
regestiration
Brief Overview of online behavioral advertising
7
Slide8Adnostic
: Privacy Preserving Targeted Audience
Key Components:
Advertiser
Publisher
Ad-Network
Tracking
Brief Overview of online behavioral advertising
anysite
8
Advertiser
Slide9Adnostic
: Privacy Preserving Targeted Audience
Key Components:
Tracking
Brief Overview of online behavioral advertising
9
Slide10Adnostic
: Privacy Preserving Targeted Audience
Key Components:
Tracking
Associate an identifier
with a user across publishers and across
different visits.
3rd party https cookieIP address
User string via browser.
Brief Overview of online behavioral advertising
10
Slide11Adnostic
: Privacy Preserving Targeted Audience
Chapter 2
Problem Statement
11
Slide12Adnostic
: Privacy Preserving Targeted Audience
Problem Statement
12
Slide13Adnostic
: Privacy Preserving Targeted Audience
Problem Statement
There is a great concern that behavioral advertising in its present form infringes on user privacy.
13
Slide14Adnostic
: Privacy Preserving Targeted Audience
Chapter 3
Goal of Research
14
Slide15Adnostic
: Privacy Preserving Targeted Audience
Goal of Research
15
Slide16Adnostic
: Privacy Preserving Targeted Audience
Goal of Research
Build an cryptographic
system by which
ad-network must bill the correct advertiser without knowing which advertisement was displayed to the user.
16
Slide17Adnostic
: Privacy Preserving Targeted Audience
Chapter
3
Adnostic
Architecture: Targeting with privacy
17
Slide18Adnostic
: Privacy Preserving Targeted Audience
Adnostic
Architecture: Targeting with privacy
18
Slide19Adnostic
: Privacy Preserving Targeted Audience
Adnostic
Architecture: Targeting with privacy
19
Behavioral Profiling:
Browser
download from ad-network a fixed list of URLs and their classification
.
List can be periodically updated and the list is same for all user.
In principle website can categorize their content in http header but since there is no incentive for the publisher we cannot rely.
Differentiate better between interest and influence. (Can be better judged locally in-browser).
User Session single user across different machine.
Slide20
Adnostic
: Privacy Preserving Targeted Audience
Adnostic
Architecture: Targeting with privacy
20
Ad Insertion:
Java-script checks if
adnostic
is installed as extension on browser.
If not then ad-network loads specific ad to render.
If yes, then ad-network sends back a list of ads, Each ad has its classification
Browser extension decides which Ad is more related to user smartly. The ad is displayed.
Slide21
Adnostic
: Privacy Preserving Targeted Audience
Adnostic
Architecture: Targeting with privacy
21
Accounting:
Main challenge is to bill the correct advertiser.
Accounting in charge-per-click does not change.
For charge per impression we propose an homomorphic encryption.
The idea is that the ad-network maintains encrypted impression counter through out the billing cycle. And rely on third party to decrypt the counter.
Slide22
Adnostic
: Privacy Preserving Targeted Audience
Chapter
4
Implementation
22
Slide23Adnostic
: Privacy Preserving Targeted Audience
Implementation
23
Slide24Adnostic
: Privacy Preserving Targeted Audience
Implementation
24
User Profiling:
Each time user visits a page
adnostic
extracts meta-data and
url
.
The list of keywords is then use to retrieve interest categories and stored in Firefox tagging.
The list derived is continuously updated aggregated for enhance user profiling.
Browser extension decides which Ad is more related to user smartly. The ad is displayed.
Slide25
Adnostic
: Privacy Preserving Targeted Audience
Implementation
25
In-browser Categorization:
When
adnostic
is downloaded it comes with built-in list of categories and cosine similarity matrix.
An update to folksonomies results in an updated matrix and is distributed to user as an update to the extension.
Folksonomies are well suited, the stay up-to-date quickly evolving the ability to categorize new words.
Slide26
Adnostic
: Privacy Preserving Targeted Audience
Implementation
26
Ad-Rendering:
The Ad-network sends the following parameter along with each ad.
A list of behavioral categories.
A score representing how relevant this ad is to the page.
Any numerical parameters that the extension accepts
The browser derives a combined score for each ad based on the score sent by the ad-network and on how well the list of categories match the user’s profile.
For each ad the associative attributes are an id, the URL, height and width of banner
Finally cryptographic material required if the billing protocol is to be used.
Slide27
Adnostic
: Privacy Preserving Targeted Audience
Implementation
27
Billing and Accounting:
Although n number of ads are pushed to browser only 1 is displayed we need to charge the correct advertiser for the impressions displayed to user.
We use homomorphic cryptographic solution and use
Elgamal
encryption on elliptic curves as it is the simplest.
We assume that each ad has id and ad-network will store in its database an encrypted counter Cid the encryption works as follows.
Initialization:
At beginning of billing period the counter is initialized to 0
Ad Insertion:
ad-network chooses ‘n’ ads and send it to browser. It remembers the ids of ads chosen, and sends a
pubKey
along.
Slide28
Adnostic
: Privacy Preserving Targeted Audience
Implementation
28
Billing and Accounting:
Ad Insertion:
ad-network chooses ‘n’ ads and send it to browser. It remembers the ids of ads chosen, and sends a
pubKey
along.
The browser choose to display ad to the user. It then creates a binary vector that represent which ad was shown to user.
For Example: if ad 2 was chosen to display. Vector representation will be (0, 1, 0, 0, 0).
The browser encrypts each component of vector using
pubKey
to obtain a vector of cypher text.
Browser sends to the ad network this cypher vector along with zero knowledge proofs
If the proofs work out ad-network multiples each element with respective costs only non-zero elements will be effective. And adds it to the homomorphic counter.
Slide29
Adnostic
: Privacy Preserving Targeted Audience
Implementation
29
Billing and Accounting:
Settlement:
At the end of billing period ad-network has to charge advertiser for the impressions the ad generated.
To do so it will need to decrypt the encrypted counter associated with all the ads.
To do so it may ask a trusted third party which holds the secret key for
pubKey
to decrypt the counter at the end of each billing cycle.
Cons: The difficulty with ultra-low rate campaigns & Advertising budget.
Slide30
Adnostic
: Privacy Preserving Targeted Audience
Chapter
5
Evaluation
30
Slide31Adnostic
: Privacy Preserving Targeted Audience
Evaluation
31
Slide32Adnostic
: Privacy Preserving Targeted Audience
Ad Rendering Time
Depicts the average ad rendering time on the four websites.
SlashDot
,
ReadWriteWeb
, SecretSoftwareClub, TheRegister
Evaluation
32
Slide33Adnostic
: Privacy Preserving Targeted Audience
Page Loading Time
As expected,
adnostic
impact on page loading delay is relatively low compared to the impact on ad rendering delay.
Evaluation
33
Slide34Adnostic
: Privacy Preserving Targeted Audience
Chapter
5
Incentives offered by
Adnostic
34
Slide35Adnostic
: Privacy Preserving Targeted Audience
Incentives Offered by
Adnostic
35
Slide36Adnostic
: Privacy Preserving Targeted Audience
Incentives Offered by
Adnostic
36
We discuss numerous reason why interest of all parties are well served by using
adnostic
Privacy Conscious publishers:
providing a privacy preserving option would enable sites who care about user-privacy to serve ads.
Lower Barrier To Entry:
User behavior is analyzed at browser which has complete visibility over user behavior pattern. Will help small ad-networks place relevant ads to user.
Regulatory Compliance:
Compliance to FTC recommendations. Allows ad-network to target ads without tracking user actions.
Potentially improved user tracking:
Many users frequently change computer, terminals or OS but modern browser use syncing capability based on which
Adnostic
can provide better relevant ads across all machines
Targeting in private Browsing mode:
Session cannot be tied to the user previous activity during incognito mode.
Adnostic
continues to rate effectively
User Control via Centralized interface
Standardized audience segmentation
Slide37
Adnostic
: Privacy Preserving Targeted Audience
Chapter
6
Possible Objections
37
Slide38Adnostic
: Privacy Preserving Targeted Audience
Possible Objections
38
Slide39Adnostic
: Privacy Preserving Targeted Audience
Possible Objections
39
Network latency and bandwidth:
While
adnostic
request ad-network for the list it may only request CDN for the specific ad to be displayed. Minimizing network latency.
Effectiveness of BA
: The list might not contain any segment from the user
behaviour
patterns
Enforcement of Non-Tracking:
Adnostic
browser extension can include a user setting that uses heuristics to disable tracking to large extent.
Click Fraud:
No change in click-on-ad mechanism. It should work as is.
Fraudulent Clicks with active contents:
The enclosing page has no visibility to iframe and is prevented from clicking on elements in the frame. (Cons: flash ads not supported).
Social Engineering attack:
Rely on filtering at ad-network level. Other way is to add noise.
Slide40Adnostic
: Privacy Preserving Targeted Audience
Conclusion
40
The
Adnostic
System need not replace the existing infrastructure instead it complements it and provides Privacy to audience at the same time. We also proposed a cryptographic billing system that is likely to play a role in any proposal for privacy preserving behavioral advertising.
ADNOSTIC Privacy Preserving targeted advertising.
Download Presentation - The PPT/PDF document "ADNOSTIC Privacy Preserving targeted adv..." 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.