[To bad that Stackoverflow closed my question. Trying to find another place.]
I'm currently doing some topic modeling. What I already found out is that what computer scientists are saying you should be doing is not necessarily what social scientists consider appropriate for TM. One thing is: weighting a document-feature-matrix using TF-IDF or not before using it for topic modeling.
I found both Ken Benoit and Julia Silge saying "don't do it". But I'm missing an explanation why you should not weight the dfm before the topic modeling. What I understand is that weighting the dfm can help to adjust to the fact that some words appear more frequently than others. So you might use it to explore your data or identify important terms in your topics. But especially CS are also using tf-idf weighting on the dfm before topic modeling.
So,
- should I weight the dfm before topic modeling, and
- if not, why not?
I'd be happy to here your say!