[Topic-models] Intuition behind CTM and DTM

Lei Tang L.Tang at asu.edu
Mon Nov 24 02:14:11 EST 2008


Hi, I have two questions about the model of CTM and DTM.

1. In correlated topic models, the topic proportion is sampled from a
logistic normal distribution instead of Dirichlet as in LDA. I didn't quite
understand the intuition behind such a modeling. Why is logistic normal
distribution has such power?

2. In both correlated topic models and dynamic topic models, the natural
parameter of a distribution is considered. Is this because of computational
convenience (for easy approximation) or any other reason?  Or is this a
standard procedure for all exponential family distributions?

Any comment is appreciated!

Thanks!
-Lei

-- 
Lei Tang
Dept.of Computer Science and Engineering,
Arizona State University
http://www.public.asu.edu/~ltang9/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/topic-models/attachments/20081124/1bf44be7/attachment.html>


More information about the Topic-models mailing list