Cross-Domain Topic Classification for Political Texts

Abstract We introduce and assess the use of supervised learning in cross-domain topic classification. In this approach, an algorithm learns to classify topics a labeled source corpus then extrapolates unlabeled target from another domain. The ability existing training data makes method significantly more efficient than within-domain learning. It also has three advantages over unsupervised models: can be specifically targeted research question resulting are easier validate interpret. demonstrate using case party platforms (source corpus) parliamentary speeches (target corpus). addition standard error metrics, we further performance by labeling subset target-corpus documents. find that classifier accurately assigns speeches, although accuracy varies substantially topic. propose tools diagnosing To illustrate usefulness method, present two studies on how electoral rules gender parliamentarians influence choice speech topics.