For an experiment with topic models, I have calculated four coherence values using Gensim's implementation:

  1. c_v
  2. u_mass
  3. c_uci
  4. c_npmi

From this paper, I know that c_v correlates mostly with human interpretation. For this reason, this seems to be the best score to use for topic evaluation. However, are there arguments for using the other measures? And how can these values be interpreted? They seem to be in a different range.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.