Beta, the percentages on the horizontal axis, represents topic-word density—with a high beta, topics are made up of most of the words in the corpus, and with a low beta they consist of few words.