[Topic-models] some questions about LDA smoothing from China

Shuangyin Li lihuihuiyin at 163.com
Fri Jul 20 03:23:09 EDT 2012


Dear All,

I am a student at Sun Yat-Sen university in China. And I have the honor to
read one of  your papers about LDA, (Latent_Dirichlet_allocation 2003). I am
very interested in it, and want to do some deep research on it. 

However, there is one problem bothering me for a long time. In LDA paper,
for the problems of sparsity come from the large vocabulary size, which a
new document is very likely to contain words that did not appear in any of
the documents in a training corpus, the paper points that the standard
approach to coping with this problem is to "smooth" the multinomial
parameters. So, I try to derive the eta(η), lambda(λ), and phi(φ) in
variational EM, but the results i get does not work just like before
smoothing! Maybe I made some mistakes.

So, who can show me the process of the inference and parameter estimation
about smoothing, and the final formulas?

Thanks a lot.

Yours,

Shuangyin Li

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/topic-models/attachments/20120720/4995b271/attachment.html>


More information about the Topic-models mailing list