My BS-meter went way up, when I scrolled down and saw the charts that conveniently "verified" the sentiment on Hacker News. I hope they don't think I typed this message, because I am depressed and lack compassion.
I would be hugely surprised if that actually gave reliable results! Human languages are hugely nuanced and context is a key part of interpreting mood in text, and even humans fail often at that. People often express surprise, for example, at pg's personality in person or in televised events, because his comments often seem (as one person put it) "Vulcan".
I'm starting to try to think of ways to thoroughly debunk their method, mostly because I'm curious if it would actually stand up.
The bar charts in the article would have been much easier to read if the horizontal axis would start at "typical". That way low would be seen as negative, while high would be seen as positive.
"presumably with the proper voting rings in place" -- [citation, etc]