It's a good start, and I dream of the day when each comment on one of my articles is a real one.
However I despair when I think about it for a bit, I feel like it may be trying to catch smoke, because of course people can (and will) change their habits.
I have had two or three line comments that have blatantly been spam. Telling me how much I have taught them and how wonderful I am, etc. etc.
The problem with these is, not only are they long, they express what could be genuine sentiment . . . hmm, maybe another part of the debate.
As far as short comments are concerned, I think we somehow need to weigh up how many genuine commenters will end up being punished because of this algo.
For instance I just had a particular short comment on my latest article, and the person seems to be genuine and leaves varying lengths of comments on people's articles, he also engages from time to time.
Perhaps the algo can be made to look at individual's average length of comment, and to check for repetition. Seeing as somebody who just posts "wow" on every single article they come across, can be thought of as a spammer.
Whereas somebody with poor English who is just trying to engage, may also leave short comments, however at least they are attempting genuine contact.
It's definitely tricky, I'm sure what I've asked isn't as easy as typing it out!