Top subset retrieval on large collections using sorted indices