-
Notifications
You must be signed in to change notification settings - Fork 198
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
What should happen if a site disagrees with the topics assigned to it by the browser? #2
Comments
Or a site could take pages that are causing its topics to be misidentified and put them in their own section: #17. That way the classifier can still classify them, but their off-topic topics don't affect the top-level topics for the site. |
Yeah a topic blocklist would be very useful I think. |
gi |
Mislabeled topics can degrade the quality of the user's topics on other sites, and we wish to iteratively improve on the classifier, but providing overrides via the platform seems to open more problems (e.g., intentionally incorrect topics) than it's worth. Note that mislabeled topics should not bring harm to the mislabeled site itself, as the site can provide its own contextual information to buyers. |
Patterns of repeatable mislabeling -- which can be recognized and trained on by server-side ML but not by individual users without significant cooperative research -- are examples of Topics API working normally in cooperation with server-side ad decisioning. The mislabeling is not a priority to fix in the browser because a known pattern of mislabeling can be trained around on a server that has sufficient data. Topics API is a way for a browser to pass an obfuscated cohort identifier to a classifier on a server -- it's essentially FLoC, but the mystery numbers are in base 629 instead of base 10. An individual user can't understand what Topics API cohort Google's ML has assigned you to just based on the individual topics in the set -- that's like saying you know what your FLoC cohort means because you know the digits 0-9. Without access to the server it's hard to say whether mislabeling different sites about the same topic differently actually provides more info to the server-side logic about the user's cohort membership than correct labeling would. Allowing sites to correct their topics at arbitrary times could make server-side ML less useful. |
👍👍👍👍👍👍👍❤️
👍
…On Fri, 6 Sep 2024, 8:57 pm StacySager77, ***@***.***> wrote:
gi
—
Reply to this email directly, view it on GitHub
<#2 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/BK5YC4PHX3O2OFNQQNR6E33ZVHGFNAVCNFSM6AAAAABNY3LTZGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMZUGM3DGNJTGU>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
Should there be a process to alter the assignment? What should the process be?
An alternative is to allow for sites to set their own topics via response header, as in #1.
The text was updated successfully, but these errors were encountered: