Efficient Similarity Search on Vector Sets