I've experimented with machine learning and haven't had great results especially if you want a small summary.

I wrote Classifier4J, and the summarizer in that does pretty well - well enough that it has been ported to C# and as a patch for Python NLTK.

Of course, maybe if the summaries were just a little better..it'd revolutionize the whole web! :-)

This doesn't really surprise me as in the time he has been working in ai/ml/nlp I don't think there would have been much time to do more than learn up to and implement some of the current state of the art. The named entity recognition for the keywords seems okay.

I did a bit of reading on summarization during my thesis, enough to know it is a big and difficult problem. I hope he keeps at it, seems like a good start but a long way to go.

However, still not sure if it ensures that i wouldn't miss out on anything important which may be fatal if it was an important email.

PLUS it doesnt say if it picks up on emotional context which is super important in interpersonal communications?