I think the premise is ambiguous and my initial reading caught the implication from "tweets" that the datasets are huge, therefore keeping track of state in the naive solution is impractical.