Characterizing large text corpora using a maximum variation sampling genetic algorithm