How does the alpha and beta hyperparameter affect the topic generation process?

Blei (2003) explained that the p(w|z) is conditioned by the beta prior and alpha has something to do with how the k-dimensionable random variable theta in the inference p(theta|alpha).

How does the alpha and beta hyperparameter affect the topic generation process?

what happens to the topic distribution when alpha is high or low?
what happens to the topic distribution when beta is high or low?

asked Apr 05 '13 at 07:45

2er0
1●4●4●6

This looks like a homework question.

(Apr 05 '13 at 08:30) Alexandre Passos ♦

One Answer:

oldest newest most voted

I won't give you the answer because it looks homeworkish, but I'll give you hints.

Look for a Dirichlet distribution with parameter alpha, and see what happens when you change that parameter.

You'll have to do some experimenting on your own, but you'll get a better sense of what's happening, also is not so hard, in Python and Matlab there are functions that already do that

answered Apr 05 '13 at 14:16

Leon Palafox ♦
4085●71●94●128

Subscription:

Once you sign in you will be able to subscribe for any updates here

Tags:

dimensionality-reduction ×31
hyperparameters ×16
latent-variables ×8
latent-dirichlet-allocation ×5

Asked: Apr 05 '13 at 07:45

Seen: 1,001 times

Last updated: Apr 07 '13 at 01:33

How does the alpha and beta hyperparameter affect the topic generation process?

Subscription:

Related questions