Exercises

• Perform k-means clustering on newsgroups data using different values of k, or use the Elbow method to find the optimal one. See if you get better grouping results.
• Try different numbers of topics, in either NMF or LDA, and see which one produces more meaningful topics in the end. It should be a fun exercise.
• Can you experiment with NMF or LDA on the entire 20 groups of newsgroups data? Are the resulting topics full of noise or gems?