如何初始化gensim LDA主题模型？

Question

It has been suggested that initializing a topic model using clusters of words can lead to higher quality models or more robust (consistent) inference. I am talking about initializing the optimizer, not setting a prior. Here is some code to illustrate what I want to do:

Create an LdaModel object, but don't pass in a corpus.

lda_model =
LdaModel(
         id2word=id2word,
         num_topics=30,
         eval_every=10,
         pass=40,
         iterations=5000)

Next assign some property of the object, corresponding to the probabilities of drawing each word from a topic to a matrix of my own construction.

lda_model.topics = my_topic_mat

Then fit the corpus:

lda_model.update(corpus)

Thanks for the help!

Answer 1

In practice, setting a prior may be a better choice than initializing the optimizer.

There are two hyperparameters alpha and eta , where alpha is a prior for the document-topic matrix and eta is a prior for the topic-word matrix. To influence word probabilities in topics, try passing eta as an additional argument:

lda_model = gensim.models.ldamodel.LdaModel(num_topics=30, id2word=id2word, eta=your_topic_mat, 
                                            eval_every=10, iterations=5000)

From the gensim docs :

eta can be a scalar for a symmetric prior over topic/word distributions, or a vector of shape num_words, which can be used to impose (user defined) asymmetric priors over the word distribution. It also supports the special value 'auto', which learns an asymmetric prior over words directly from your data. eta can also be a matrix of shape num_topics x num_words, which can be used to impose asymmetric priors over the word distribution on a per-topic basis (can not be learned from data).