In the last year this blog has sort of languished.
I have been busy giving talks and doing personal projects.
One of the most interesting things that I’ve had the pleasure of doing is giving a talk on some NLP work that I did in my free time. In particular, I talked about using computational political science to try to classify arbitrary news articles as politically biased using a variety of techniques.
In particular, I found this great dataset of political speeches from the members of the 111th congress. Using an ideal-point mapping onto the real spectrum from voteview.com, I had an ample set of training data and a model with some statistical soundness to use to classify those documents into political buckets.
From this it’s an easy to use a variety of supervised learning techniques to map arbitrary text onto a political spectrum. The wonderful people at Kent video taped the whole talk so everyone can see just how poor of a public speaker I am.