[Topic-models] Training Classifier with multi-labeled data

risaN yamaR nasirramay at gmail.com
Thu Mar 4 06:31:37 EST 2010


Hi All,
I want to use LDA as base model for web 2.0 content analysis.
I am searching for some objective method to decide the values for hyper
parameters (i.e. alpha & beta) and deciding the number of topics for the
collection.
Second thing is, if there is a way to decide objectively about the quality
of topics discovered by LDA?
My third question is related to compute the distance between topics, which
one is a good measure for this, JS divergence or KL divergence ? or some
other measure ...

thanks a lot in advance,

Nasir Naveed
Uni Koblenz
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/topic-models/attachments/20100304/bb786d46/attachment.html>


More information about the Topic-models mailing list