How do I get topic numbers in LDA model in gensim

1
votes

I have trained LDA model using gensim on a text_corpus.

>lda_model = gensim.models.ldamodel.LdaModel(text_corpus, 10)

Now if a new text document text_sparse_vector has to be inferred I have to do 

>lda_model[text_sparse_vector]
[(0, 0.036479568280206563), (3, 0.053828073308160099), (7, 0.021936618544365804), (11, 0.017499953446152686), (15, 0.010153090454090822), (16, 0.35967516223499041), (19, 0.098570351997275749), (26, 0.068550060242800928), (27, 0.08371562828754453), (28, 0.14110945630261607), (29, 0.089938130046832571)]

But how do I get the word distribution for each of the corresponding topics. For example, How do I know top 20 words for topic number 16 ?

The class gensim.models.ldamodel.LdaModel has method called show_topics(topics=10, topn=10, log=False, formatted=True), but the as the documentation says it shows randomly selected list of topics.

Is there a way to link or print I can map the inferred topic numbers to word distributions ?

3
pythonnlpldagensim
You can use show_topic(). See here: github.com/piskvorky/gensim/blob/develop/gensim/models/… I think the ordering is just arbitrary but I could be wrong. What exactly do you want to do? - Karsten

3 Answers

6
votes
lda.print_topic(x, topn=20)

will get you the top 20 features for topic x

0
votes

Or if you have K topics, then:

print(str(["Topic #"+str(k)+":\n" + str(lda.show_topic(k,topn=20)) for k in range(K)]))

will get you ugly, but consistently sorted output.

0
votes

The last line here will change the number of words per topic. Hope this helps :)

# train and save LDA model

lda_model = gensim.models.LdaMulticore(bow_corpus, num_topics=20, id2word=dictionary, passes=2, workers=2, chunksize=400000)

# check out the topics 

for idx, topic in lda_model.print_topics(-1):
   print('Topic: {} \nWords: {}'.format(idx, topic))

# swap out '30' for any number and this line will give you that many words per topic :)
lda_model.print_topics(idx, 30)