A discussion about implementing an anti-gibberish filter for post bodies in Discourse. Bro proposed the idea, citing the existence of a similar feature for topic titles that checks for entropy. Bro suggested that this check could be applied to post bodies, as human-typed gibberish often exhibits patterns and low entropy. However, codinghorror expressed concerns, stating that gibberish posts are rare and that titles are more important to protect, as they are visible on the home page. elberet supported Bro’s idea, highlighting that human-typed gibberish is not truly random and can be detected using entropy checks.
Hello Jeff. Thank you and your colleagues for such a great piece of software.
I am not a computer scientist, so please excuse me if I stick my foot in it. According to Wikipedia, an English text (and I’m guessing that the same goes for most other western languages) would typically have a low entropy, as opposed to gibberish, so is there any reason why the same check run on titles cannot be run for the body of a post? I think this would also avoid to some degree people just typing out random stuff to fill the character quota mentioned in the original thread, right?
Like the mythical ALL CAPS POST, posts consisting of all gibberish are pretty rare in the wild. If we saw posts like this daily then it would be a more serious concern.
Titles are more important to protect as what you see on the home page is titles, not post bodies.