O'Reilly logo
  • Qi Fan thinks this is interesting:

Now, given a set of items with their associated words as the observed data,

From

Cover of Statistical Methods for Recommender Systems

Note

Symmetrically:
Alpha represents document-topic density - with a higher alpha, documents are made up of more topics, and with lower alpha, documents contain fewer topics.
Beta 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.

Asymmetric:
higher alpha results in a more specific topic distribution per document. Likewise, beta results in a more specific word distribution per topic.