Perhaps adding a way for users to tag a story as left or right leaning (or even more descriptive tags like anti-union, pro-life, anti-regulatory etc) might also provide some insight to your classifiers. How an individual classifies a story would only need to be visible to that user, but from a UX perspective it might encourage more honest input into the system that could be used to ground political viewpoints. You might find that a lot of partisans mark themselves as neutral, but most partisans won't be able to resist labeling a story in a manner that fits their partisan perspective.