Toyota AI Seminar: Classifying the Political Leaning of News Articles and Users From User Votes

A graph random walk model propagates classifications of political news articles and users as conservative or liberal. Using data from a large news aggregator, Digg.com, we view a user’s “digg” on a news article as an approval vote, and assume that liberal users will digg liberal articles more often, and similarly for conservative users and articles. Starting from a few labeled articles and users, the random walk algorithm propagates political leaning labels to the entire graph. It achieves 95.4% accuracy on articles and 99.5% accuracy on users. The random walk model, using the subjective liking data from users, performs much better than a traditional SVM text classifier, which achieved 76.3% accuracy on items.
Daniel Xiaodan Zhou is a PhD student at the School of Information. His dissertation is about using computational approaches to study and solve social problems, especially in the domain of social media. He is also working on recommender systems and the study of Chinese political online discourse.