Comment by sonium

I did something similar (but not for documents) but I’m struggling with selecting the optimal number of clusters.