0

When we visualize the LDA using pyLDAvis, we can see topic overlap. I want know the word that is causing this topic overlap. Or I want to know the words that are at the intersection of the topic bubbles. Any guidance is appreciated.

Sid
  • 552
  • 6
  • 21
  • By "seeing topic overlap", do you mean that projections of topics have an intersection e.g. topics 2 and 5 in this image? https://miro.medium.com/max/4832/1*k0svin9aq0ObRnHZgg6mFQ.png – arnaud Apr 21 '20 at 07:54
  • Yes, that is correct. – Sid Apr 21 '20 at 10:54

1 Answers1

0
  1. Select topic 2
  2. Copy a list of words from the right (the ones with bars)
  3. Select topic 5
  4. Copy the list as in 2.
  5. Compare the copied lists on your own.

I am not sure there is any better method there...

sophros
  • 14,672
  • 11
  • 46
  • 75
  • 1
    Without using the interface (and copy/pasting) you could also imagining just doing the same with topics word distribution. Comparing top-100 words per topic... – arnaud Apr 21 '20 at 13:20
  • isnt it possible that top 100 words might not have the overlap... – Sid Apr 24 '20 at 18:21
  • Depending on the training parameters but I would assume that there should be overlap among top 100 word sets if the topics overlap. – sophros Apr 24 '20 at 18:25