Top-k queries on uncertain data: on score distribution and typical answers