Facebook announces new system ``PETs'' for posting effective advertisements while maintaining user privacy
Apple has revised its privacy policy from iOS 14 and launched `
Privacy-Enhancing Technologies and Building for the Future | Facebook for Business
https://www.facebook.com/business/news/building-for-the-future
What Are Privacy-Enhancing Technologies (PETs) and How Will They Apply to Ads? - About Facebook
https://about.fb.com/news/2021/08/privacy-enhancing-technologies-and-ads/
In light of the fact that Apple and Google continue to make privacy changes through their own web browsers and operating systems, and that privacy-related regulations continue to be tightened, Facebook said, ``Digital advertising is individual third-party data. It is important to recognize that the time has come to evolve in order to reduce our dependence on it,' he said, announcing 'PETs' as a technology for that purpose.
“We believe that personalization in marketing is the best possible experience for people and businesses,” Facebook said. , The cost itself will be higher, and people will see ads that are not relevant, timely, and uninteresting, ”he emphasizes the importance of targeted advertising .
Facebook said, “We are optimistic that PETs will prove that the industry can continue to serve targeted advertising even as we reduce our reliance on individual third-party data. It is useful for displaying relevant advertisements to people and measuring the effectiveness of advertisers' advertisements, while keeping it to a minimum,' he said. I appealed that it would be a technology that can
PETs are supported by three technologies: '
◆MPC
MPC, a privacy-enhancing technology, ``limits the information that one can learn by linking two or more organizations''. Data is end-to-end encrypted so that neither party can see the other's data during transfer, storage and use, protecting user privacy while allowing multiple parties to will be able to measure advertising effectiveness. Specifically, if one organization has 'information about users who saw ads' and another organization has 'information about who bought what', each organization has each other's dataset It will be possible to measure only the effect of posting advertisements without disclosing.
In addition, Facebook has already started testing the MPC-based advertising effectiveness measurement tool `` Private Lift Measurement '' in 2020, and plans to provide the same function to Facebook advertisers in 2022. In addition, Facebook has released the framework of this solution as open source as 'Facebook Private Computation Framework (FBPCF)' so that anyone in the industry can develop advertising effectiveness measurement tools based on similar technology. increase.
GitHub - facebookresearch/fbpcf
https://github.com/facebookresearch/fbpcf
◆ On-device learning
Also, in PETs, data such as ``which user bought what'' is processed locally on the terminal instead of being processed on a remote server or cloud, and the algorithm is trained. This makes it possible to display targeted advertisements suitable for each user without knowing what actions individuals are taking on apps and websites.
For example, if people who click on ads for exercise equipment are more likely to buy protein, on-device learning identifies just that pattern without sending individual data to Facebook servers or the cloud. becomes possible. Then, Facebook uses the pattern derived from on-device learning that ``many people who click on exercise equipment ads tend to buy protein'' to display protein ads to suitable users.
Since the accuracy of on-device learning improves over time, ``As time passes, the accuracy of targeting advertisements will increase, and the number of cases in which less relevant advertisements will be displayed will decrease,'' Facebook claims. doing.
◆ Differential privacy
Differential privacy is a technology for protecting data from re-identification, preserving privacy by including carefully calculated 'noise' in the data set. For example, if 118 people clicked on an ad and then purchased a product, the differential privacy system would add or subtract a random amount from that number. In other words, instead of the number '118', numbers such as '120' and '114' are output to hide the exact number.
By creating such 'noise', it is very difficult to identify who purchased the product after clicking the advertisement. In addition, it seems that differential privacy is frequently used in large-scale data sets released in public research.
In addition, Facebook believes that collaboration within the industry is essential for the success of these advertising-related tools, including the Partnership for Responsible Addressable Media (PRAM), the World Wide Web Consortium (W3C), It reveals that it is working with industry groups such as the World Federation of Advertisers (WFA) to improve the tool.
In addition, when the technology media The Verge interviewed Mr. Graham Mudd, Facebook's vice president of product marketing, Mr. Mudd said, ``PETs are very meaningful participants on forums like W3C. It's the same kind of technology that Google seeks feedback from and discusses with FLoC .' 'FLoC addresses a specific use case (behavioral targeting) without revealing anything about a specific individual. And we is currently running a beta version of PETs, which is also measurement-focused and not specific-individual-focused: in some cases the FLoC approach is appropriate, in some cases the PETs approach is In some cases, these technologies do not necessarily compete with each other, ”he explains the relationship with Google's FLoC.
FLoC is an API devised by Google, which is trying to build a new advertising mechanism without third-party cookies, and is a mechanism for posting effective advertisements while considering user privacy in the same way as Facebook's PETs. . This FLoC has been criticized by web browser Brave, software development company Oracle, and the Electronic Frontier Foundation.
Why is 'FLoC' under development by Google 'harmful' and what is the damage caused to users and websites? -GIGAZINE
Related Posts:
in Web Service, Posted by logu_ii