I'm looking to try and use deep learning methods for topic modeling as opposed to the more traditional methods of lda and word embedding methods. However, I'm having trouble finding good labeled datasets for this task. So far the best that I've seen is the New York Times Dataset which I can't use due to licensing constraints. I've also seen the 20News Dataset but it only has twenty categories so it probably won't scale well to other domains.

Are there any other good datasets out there that I'm missing that can be used for topic modeling? I'm happy to use a dataset that isn't explicitly meant for topic modeling; as long as it has some sentences/paragraphs that are tagged or labeled that should be fine.

| improve this question | | | | |
  • $\begingroup$ It seems to me this question is better suited for open data. $\endgroup$ – wacax Nov 26 '19 at 21:28
  • $\begingroup$ @wacax I just posted it there as well at opendata.stackexchange.com/questions/15880/…, thanks. It seems like this is a more active site so hopefully I can leave it up here as well. $\endgroup$ – David Nov 26 '19 at 21:42

Since topic modeling is unsupervised, it's not usually evaluated against labeled data. Instead people devise measures which evaluate the clusters, typically based on how much the most probable words for a topic are semantically relevant.

You might find data ideas in this paper: https://www.cs.cornell.edu/~laurejt/papers/authorless-tms-2018.pdf

| improve this answer | | | | |
  • $\begingroup$ Thanks for the reference. I believe that if I wanted to use a deep learning model (some kind of seq2seq/rnn model) on it, then it would need to be a supervised method, unless I'm mistaken. $\endgroup$ – David Nov 27 '19 at 17:11
  • $\begingroup$ As far as I know the term "topic modeling" refers to unsupervised topics, so what you're looking for is probably more text classification datasets. $\endgroup$ – Erwan Nov 27 '19 at 18:40

Not the answer you're looking for? Browse other questions tagged or ask your own question.