I'm using VADER - https://github.com/cjhutto/vaderSentiment - because it's trained on NYT data which makes it suitable for news sentiment parsing.

The code is pretty readable but relies heavily on a ruleset which might need to be tweaked for one's need.