Finding and approximating top-k answers in keyword proximity search