search

akashgit / autoencoding_vi_for_topic_models

Tensorflow implementation for prodLDA and NVLDA.
http://openreview.net/forum?id=BybtVK9lg
MIT License
252 stars 51 forks source link

How to get the average topic coherence? #4

Closed hilda92 closed 7 years ago

hilda92 commented 7 years ago

Hello, Akash ! Thanks for sharing the code!!! I have some problem about the average topic coherence. In paper, the average topic coherence is NPMI between all the pairs of words in a set of topics. If all the pairs of words in a set of topics means the top n words of each topic of beta? If beta is K*V, NPMI=sum( NPMI(x,y),x y is top n words of each topic of beta) ) / K ? Can you share the code about the computing average topic coherence? Help wanted! Thank you!

akashgit commented 7 years ago

Hi, the topic coherence code is avaliable from the original author's of that paper. Also you can look up https://github.com/hyqneuron/pytorch-avitm . @hyqneuron has his own implementation for a pytorch version of AVITM.

nzw0301 commented 7 years ago

Maybe unrelated my comment, I also release my notebook for a Keras version of your prodLDA model. I'm happy if someone uses my code. https://github.com/nzw0301/keras-examples/blob/master/prodLDA.ipynb

akashgit commented 7 years ago

Thanks @nzw0301 If you don't mind I'd like to link your notebook on the project's readme?

nzw0301 commented 7 years ago

@akashgit Of course!