The gensim [0] package has a nice implementation of online LDA that can handle massive streaming datasets. If you want to avoid specifying the number of topics, you can use HDP-LDA. David Blei (inventor of LDA) has a reference implementation on his website [1], along with many other variants of LDA.