Quora Feed Personalization: Choosing Signals and Objectives for Recommendations

You are a product manager at Quora. Explain how you would decide which content Quora should recommend to each user: what the recommendations should achieve, which signals should drive them, how those signals should be weighed against each other, and how you would tell whether the recommendations are working. ```hint Two sides of every recommendation Each recommendation decides both what a reader sees and which writer gets an audience. Define what a good recommendation means for both groups before you choose signals. ``` ### Clarifying Questions - Which surface is this for: the home feed, the email digest, related questions on an answer page, or all of them? - Does "content" mean answers to read, questions for experts to answer, or also topics, Spaces and people to follow? - Is a recommendation system already in place that we are improving, or are we designing the approach from scratch? - Should the recommendations also serve logged-out visitors who arrive from search engines, or only signed-in members? ### What a Strong Answer Covers - A goal for recommendations that serves readers, writers and Quora's business, and a definition of a good recommendation that goes beyond clicks. - Segments with different needs and different amounts of data (new versus established members, readers versus writers), including cold start for new members and new answers. - The signals to use (stated interests, behavior, follows, answer quality, freshness, context) and how they are prioritized and combined, with room for variety and discovery. - Quality and safety filters for low-quality or misleading answers, clickbait and sensitive content, and limits on what personal data is used. - Success metrics with guardrails, and a plan for testing changes. ### Follow-up Questions - A ranking change raises clicks from the email digest, but downvotes and unsubscribes also rise. What do you do? - How would you give a strong answer from a new writer with no followers a fair chance to be seen? - Should a member's activity outside Quora be used to personalize recommendations? Where would you draw the line?