This c++ project provides a multithreaded BTM CLI application. See the python file for an example for how to use it.
Specifically, in BTM we learn the topics by directly modeling the generation of word co-occurrence patterns (i.e. biterms) in the whole corpus. The major advantages of BTM are that 1) BTM explicitly models the word co-occurrence patterns to enhance the topic learning; and 2) BTM uses the aggregated patterns in the whole corpus for learning topics to solve the problem of sparse word co-occurrence patterns at document-level.
The key idea for parallelizing the sampler in the multicore setting is that the global topic distribution and the topic-word table change only little given the changes in a single document. Hence, we can assume that n(t) and n(t,w) are essentially constant while sampling topics for a document.