[Topic-models] Reconstructing the input matrix in LDA

Basma Alharbi basma.harbi at gmail.com
Thu Mar 17 04:21:22 EDT 2016


Hello all,

I would like to know if it is reasonable to reconstruct the original input
(bag-of-words) from the learned parameters (phi and theta) in LDA?.

I understand that I will not get frequencies in the reconstructed matrix.
But, let us say that i normalize the bag-of-words matrix and the
reconstructed matrix, is it valid to compare different topic models based
on how well they 'reconstruct' the original input space (e.g., computing
the reconstruction error using cross entropy) ?

Best,
Basma
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/topic-models/attachments/20160317/3f5b429a/attachment.html>


More information about the Topic-models mailing list