I'm building LDA topic models in to apply against a collection of small texts and regardless of the number of topics, I'm finding that there is always one topic that is very large (in terms of representation) and the rest are very small;
This is a histogram of the strength of topics within the texts;
I'm not sure whether it's my settings of the LDA or not, there aren't any obvious articles online of people having a similar problem.
These are the parameters I'm feeding;
lda_model = gensim.models.LdaMulticore(bow_corpus, num_topics=clusters, id2word=dictionary, passes=400, alpha = 1.0/clusters, workers=4)
Could it also be the size of the texts? This is a histogram (with log_y) of the distribution of the number of words in the text;