Do Author-Topic Models Work With PyLDAvis

[john.hard...@gmail.com]

Basic Question. If I train an author-topic model with gensim, can I then visualize it with pyldavis? Are there any parameters I ned to change to get it to work.

Harry

[Ivan Menshikh]

Hi John,

I think olavurmortensen and parulsethi can help you.

[parul sethi]

It should be possible as the gensim's LDA APIs that are required for pyLDAvis are all valid for author-topic model too. But you'll only be able to visualize topic-term relationships (as in usual LDA), and nothing related to Author info.

[john.hard...@gmail.com]

Great, that's what I was wondering about. I'd like to explore other topic models for the project I'm working on, but pyLDAvis is such a useful tool.

This is unrelated to my original question, but do you happen to know if author metadata provides semantic information to topics created, or does it just allow you to find the topic distribution of authors? I know this is something that I need to experiment with, but I'm wondering if there's anything in the math behind it that makes it's topics 'better' than pure LDA.

[Ólavur Mortensen]

To your unrelated question: the author-topic model is more flexible than LDA, so in theory it should be able to fit the data better. Whether better model fit leads to better topics is unclear.

From a very anecdotal standpoint, when I was working with the author-topic model I was constantly comparing with LDA, and generally found the author-topic model to produce more intuitive topics. Don't quote me on this though, test it for yourself :)

[john.hard...@gmail.com]

I definitely will. I'm working with a dataset of grant applications that belong to specific departments that I want to tag as meta data. Topics cut across departments, but I imagine the fact there are already somewhat pre-defined categories should be significant.