I did the same!

It also doesn't seem to like swear words and web abbreviations (like j/k for example). Perhaps with more data (from twitter or similar) it could go a long way though; as it definitely needs to learn about the more uncouth words.

Thwarted expectations and sarcasm are difficult to handle. You are down to the choice of optimizing for this particular case or the majority of text.

You've found the 15% it gets wrong. Though, given the other tricky cases it does well on, more training data might help.

Borat is not a measure by which any sentiment analysis should be done... :P