Topic Models_ from reviews


This is a companion discussion topic for the original entry at https://kni.me/w/zXa_WBQgRZz4nqq6

Hello Francisco,

I find your solution for topic number determination (Block 2) really interesting. Is there a more detailed explanation about how to perform this two-step perplexity method?

For me it is not clear how broad should be the range in the Step 2. In the meta-node description, it says the range of topics is [2,20] but the range in the resulting chart is [13, 17]. Thank you!

Regards,
Max

In the blog post we used the range [13, 17] for the number of topics. Initially we used [2, 20] but we got the important results in the range [13, 17]. So, to speed up the execution of this workflow we concentrated on this range. However, the original one was [2, 20]. Outside of this range the perplexity plot is not so interesting. Of course we did not know till we tried. I hope this answers your question,