This Topic Modling on Non Matrix Factorization is about the CNN news and how overly utilizing CNN as a new platform where the most topics is being vocalized.
In My Model:
From the 30 best topics the highest coherence score (.435) There has been a peak and drop session observed.
10 topics was a close second in terms of coherence score (.432) so you can see that that could have also been selected with a different set of parameters. So, like I said, this isn’t a perfect solution as that’s a pretty wide range but it’s pretty obvious from the graph that topics between 10 to 40 will produce good results. That said, you may want to average the top 5 topic numbers, take the middle topic number in the top 5 etc. For now we’ll just go with 30.