I call in Python gensim library to train the LDA, trial and error in different data sets and different number of keywords, found that trained at a time difference between the results of the subject is not big, back and forth is that a few keywords, there was no other words, is this why ah
