Facebook Announces Privacy Policy and Targeted Advertising Posting System

Apple revised its privacy policy in iOS 14 and introduced App Tracking Transparency (ATT), a measure to prevent advertising companies from collecting advertising identifier IDFAs without notifying users. Facebook, which strongly opposed this, warns that advertising revenue will drop sharply due to the difficulty of targeted advertising, and is fighting. There is also a report that the revenue of advertisers and app developers who serve ads on iOS through Apple ATT decreased by 15-20%.

In response, Facebook announced Privacy-Enhancing Technologies (PETs), a new technology that effectively publishes advertisements while protecting user privacy without relying on existing advertising identifiers such as IDFA.

Based on the fact that Apple and Google are changing their views on privacy through their own web browsers and operating systems, and regulations related to personal information protection are being strengthened and continued, it is time for digital advertising to evolve to reduce reliance on individual third-party data. PETs were announced as a technology for this, saying that it is important to recognize that there is

Facebook says that individual is the best experience possible for people and businesses in marketing. Without personalized advertising, starting and growing a business is difficult, making it difficult to find new products and services. It emphasizes the importance of targeted advertising.

Facebook is optimistic that PETs can demonstrate that the industry can continue targeted advertising while reducing reliance on individual third-party data. PETs minimizes and suppresses the amount of personal information processed and shows relevant advertisements to people to help measure the advertising effectiveness of advertisers. there is.

There are three things that support PETs: Secure Multi-Party Computation (MPC), Warm Machine Learning, and Differential Privacy Protection.

MPC, a technology for enhancing personal information protection, limits the information that one can learn by linking two or more groups. Because data is end-to-end encrypted, when transmitted, stored, or used, no party can see the other party’s data. This allows multiple parties to measure the effectiveness of this advertisement while protecting user privacy. Specifically, if an organization has information about users who have viewed an advertisement and another organization has information about who has purchased what, and each organization has published an advertisement without disclosing data sets to each other, only the effect can be measured.

In addition, Facebook has already begun testing its MPC-based advertising effectiveness measurement tool in 2020, and will provide features for Facebook advertisers in 2022. In addition, Facebook has released its solution framework as an open source FBPCF (Facebook Private Computation Framework) so that all industries can develop advertising effectiveness measurement tools based on the same technology.

Next is in-device learning. PETs train algorithms to process data locally, such as which user bought what, on a terminal rather than on a remote server or in the cloud. This makes it possible to display targeted advertisements suitable for each user without having to know what action the individual is taking on the application or website.

For example, if most people who click on exercise equipment ads tend to buy protein, in-device learning can only identify those patterns without sending individual data to Facebook’s servers or the cloud. And Facebook uses the pattern that most people who click on exercise equipment ads, derived from in-device learning, tend to buy protein, to show protein ads to the right users.

According to Facebook, in-device learning improves over time, resulting in less relevant ads being shown as the accuracy of targeted ads increases over time.

Finally, differential privacy. Differential privacy is a technology that protects data from being re-identified. Maintain privacy by including carefully calculated noise into the dataset. For example, if 118 people click on an advertisement and purchase a product, the differential personal information system adds or subtracts a random amount from this number. In other words, it hides the exact number by outputting the number 120 or 114 instead of the number 118.

It becomes difficult to pinpoint who will buy a product after clicking on this noise-generating ad. In addition, it is said that differential personal information is frequently used in large datasets released in public research and the like.

Facebook believes that cooperation within the industry is essential for the success of such advertising tools, and is working with PRAM (Partnership for Responsible Addressable Media), W3C (World Wide Web Consortium), and WFA (World Federation of Advertisers) to improve the tool. there is.

A Facebook official said PETs are the same kind of technology as FLoC, where Google asks for feedback and discusses it in forums like W3C, where it is a meaningful participant. FLoC addresses specific behavioral targets without revealing anything about specific individuals. Beta also focused on measurements, not on specific individuals. He added that these technologies are not necessarily in conflict, as the FLoC approach may be appropriate and the PETs approach may be optimal.

FLoC is an API devised by Google to build a new advertising structure without third-party cookies. Like Facebook PETs, it is a structure to display effective advertisements while considering the protection of user privacy. Regarding FLoC, there are also voices of criticism from the web browser Brave, Oracle, and the Electronic Frontier Foundation. Related information can be found here.