Ranking Stories for a Google News Feed: Objective, Signals and Pipeline

Design the algorithm behind a news feed for Google: the logic that decides which stories the feed shows and in what order. ```hint Objective first Before listing signals, decide what a good feed means for the reader and for Google; that choice determines what the system should predict and what it must never trade away. ``` ### Clarifying Questions - Which surface is this: the feed inside the Google mobile app, the dedicated news product, or something new? - Should the feed be personalized for each signed-in user, or show the same top stories to everyone in a region? - Is the content only articles from publishers, or also videos and posts from other sources? - What matters most to Google here: time spent, how often people come back, or users leaving well informed? ### What a Strong Answer Covers - An explicit ranking objective, and why it beats the obvious alternatives - Where candidate stories come from and the families of signals used to score them - A stage-by-stage pipeline from retrieving candidates to the final ordering - Treatment of freshness, many articles about one event, diversity, new users and new stories, and low-quality or false content - Offline and experimental evaluation, including guardrail metrics ### Follow-up Questions - A major event breaks and the first reports contradict each other. How should the feed rank stories about it in the first hour? - Publishers say your change cut the traffic they get from the feed. How do you decide whether that is a problem to fix? - How do you keep personalization from narrowing what each user sees over time?