3 ms·
Combing for Insight in 10k Hacker News Posts with Text Clustering
- PaulHoule 4y agoI like it, the cluster plots are more interpretable than many I've seen, but I have a book on numerical taxonomy from the 1960s that talks about KMeans. There are much better ways to cluster and layout high-dimensional points in a low-dimensional space like https://en.wikipedia.org/wiki/T-distributed_stochastic_neighbor_embedding https://en.wikipedia.org/wiki/T-distributed_stochastic_neigh... I'm also curious how much gain you get using the embedding you use as opposed to old fashioned bag-of-words.