Closed relevance-api-topics closed 2 weeks ago
It's based on the frequency of a subset of browsing histories. A browsing history entry (a page) is eligible to participate only if the page had at least one topics caller. See when topics history storage get populated in the spec.
@xyaoinum thanks for the reply. based on the subset of the browsing histories, do you further rank topics for each caller separately?
ask this based on the understanding that different callers have different 'rights' for topics, only when callers observe the topics, they can receive 1 randomly selected from the top 5.
the 'top 5' are generated within one epoch for all the callers?
The ranking depends on the frequency, and also on a a binary priority level (see the announcement, topics-utility-buckets-v1.md, pending spec update). However, it does not depend on the frequency of callers.
But note that it doesn't mean all the callers can get all top 5 topics in the next epoch. Per the explainer: "Only callers that observed the user visit a site about the topic in question within the past three weeks can receive the topic." The browser does need to track which caller observed which top 5 topics (corresponding to the top 5 topics with caller domains in the spec).
for "Lastly, Chrome selects the top five as the user's top topics for that epoch, which are eligible to be shared with callers."
still hope to confirm: 1) the 'top five' are generated for all callers at one single ranking, and filter by each caller's user observation, or 2) chrome maintain each caller with separately ranked 'top five'?
@relevance-api-topics Your first statement is accurate
When getting 'top 5 topics', would we count frequency of websites visits based on each eligible caller, or always count based on the browser's whole visiting histories? Thanks.