Recommend Three Boards When a User Saves a Pin

When a user saves a pin, suggest the three most relevant boards that the user might want to save it to. How would you define and model this recommendation problem? ### Clarifying Questions Are candidates limited to the user's existing writable boards, or can new or shared boards also appear? Can there be fewer than three eligible boards? What pin content, board content, and prior save history are available? For a concrete baseline, use existing boards the user can write to and return up to three distinct eligible boards. These are explicit practice choices; no production scale or latency target is supplied. ### What a Strong Answer Covers Define the prediction unit and labels, construct eligible candidates, represent pins and boards, rank them, handle sparse users and boards, and evaluate both recommendation quality and the save experience. ### Follow-up Questions - How would you represent a new board with only a title and no saved pins? - What bias arises if the system learns only from saves made after recommendations were shown? - How would you handle several boards with nearly identical themes?