Open eysegal opened 11 months ago
The kind of targeting you're describing is not supported in Protected Audience today, and doing so would be a change to the underlying privacy model. There's been some recent discussion of this in https://github.com/michaelkleber/privacy-model/issues/26 opened by @thegreatfatzby.
I discussed some of this over in this comment.
all IGs are sent to the trusted server together, but I don't think that the intention is to look at them together,
Each response value for each key is meant to be based only on the key and hostname as per the trust model.
Thank you for your reply @JensenPaul Referring to "may dampen a browser’s ability to protect users from ads they find disagreeable" - I thought the purpose is not to allow ad companies to hold a lot of information about the users. If the users are able to receive more relevant ads, but without data being passed to the ad companies, then wouldn't that be a better experience for all? Personally, and I'm setting aside the fact that I currently work in an ad company, as a user, I prefer to get ads that are relevant. I just don't want anyone to hold that information on me - and that's exactly what Protected Audiences can bring.
I know it may be too naive to comment, but have we considered setting a minimum number of users in the combined IG that is bigger than 50, adhering to the privacy model? starting with maybe all/some that a seller owns?
Hi @omriariav: Each IG can hold information about a person's past behavior on a single site, and the detail in that information is unrelated to the number of different people in the IG. So the kind of k-anonymity limit that you're proposing wouldn't really have any effect — the bidder would still be able to place bids based on a person's actual cross-site activity on multiple sites.
Sometimes advertisers use a combination of user segments to target their ads. For example target audiences who read about baby products and financial products to upgrade their car. To my understanding, in the current proposal each IG is examined separately, without dependency with other IGs, so targeting would be a lot less accurate. I see that all IGs are sent to the trusted server together, but I don't think that the intention is to look at them together, but to save network calls (maybe I'm wrong).
Is there a solution for that?