PLSA with gensim

Roxana Attar

unread,

Apr 4, 2017, 4:31:40 PM4/4/17

to gensim

Hi everyone,
I am working with different topic modeling algorithms and now I wanna try PLSA. But I can't find any definition of it in gensim.However, I have found a post that if we use ldamodel with parameters alpha=1 and eta=1 it will work as PLSA. I did this and the only thing that I got as the results were topics with words but all the words' probabilities were 0 ! Has anyone got any experiment on PLSA?
Thanks.
Roxana

Lev Konstantinovskiy

unread,

Apr 4, 2017, 7:14:42 PM4/4/17

to gensim

Hi Roxana,

All the words having probability 0 is strange. What code did you use to train the model?

Roxana Attar

unread,

Apr 4, 2017, 10:05:10 PM4/4/17

to gen...@googlegroups.com

Hi Lev,

I exactly used the same LDAmodel ! the only thing that I changed was the parameters of ldamodel which I added an alpha=1 and an eta=1 !

--
You received this message because you are subscribed to a topic in the Google Groups "gensim" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/gensim/Dm5uNsyfkno/unsubscribe.
To unsubscribe from this group and all its topics, send an email to gensim+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Roxana Attar

unread,

Apr 28, 2017, 3:39:24 PM4/28/17

to gensim

Would Somebody please tell me how can I implement PLSA in gensim?
Thanks.

Lev Konstantinovskiy

unread,

May 2, 2017, 8:41:51 PM5/2/17

to gensim

Hi Roxana,

If you could post a small corpus and the exact code you are trying to run then it would be easier to help.

Setting alpha=1 and eta=1 doesn't exactly produce pLSA however it does make the Dirichlet prior dense which is more similar to pLSA.

There are existing implementations of pLSA in python. Starting with a gensim wrapper around would be a good first step.