LDA placing weights on topics

Hi everyone,

Is there a known method or previous work on how to assign weights to topics obtained from the LDA algorithm and combine them into a single weighted topic vector? I have come across the Term Frequency-Inverse Document Frequency (tf-idf) matrix, which is integrated into MATLAB but requires the use of the bagofwords() expression. I have also searched for information on UMass and CV, but it doesn't seem to be available in any of the toolboxes (please correct me if I'm wrong).

Therefore, I would be more than grateful for any recommendations or tips. Many thanks!

Rob

Answer 1

[Pranjal Saxena]

Hi Rob,

I understand that you want to assign weight to topics obtained from LDA algorithm and combine them into a single weighted topic vector.

MATLAB provides the “bagOfWords” function and the “tfidf” function in the “Text Analytics Toolbox”, which allows you to calculate tf-idf weights for a collection of documents. You can use these functions to create a tf-idf matrix and apply it to the topics obtained from LDA.

I would like to suggest you refer to the following MATLAB documentations for more information about it.

Text Analytics Toolbox Documentation (mathworks.com)

Latent Dirichlet allocation (LDA) model - MATLAB (mathworks.com)

Term Frequency–Inverse Document Frequency (tf-idf) matrix - MATLAB tfidf (mathworks.com)

I hope this helps you.

[Rob]

Hi @Pranjal Saxena,

Thanks for your answer! That's what I originally thought I would do, compute the weights via tf-idf and then apply them to the LDA outcome, until I came across this post. It's basically saying we can't combine both approaches, unless I am reading this wrong. Apologies, this might be a more data/stats question but it would be great if I could get a second opinion on this, because using the approach you described makes total sense. Thanks!

Rob