Recently I’ve been writing F# code to process and determine if a tweet is interesting based on previous tweets that I give it. It uses what’s called a Bayesian classifier to determine if a tweet goes into the interesting or boring category. I’ve learned a lot by writing this and it still doesn’t work exactly as I’d like. The problem is finding the right feature set to train the model against. I’m going to use the top 50 words and their distributions from the interesting tweets as features now and see if that improves the model’s ability to predict tweet interestingness.