Tradeoffs in Accuracy and Efficiency in Supervised Learning Methods

Words are an increasingly important source of data for social science research. Automated classification methodologies hold the promise of substantially lowering the costs of analyzing large amounts of text. In this article, we consider a number of questions of interest to prospective users of supervised learning methods, which are used to automatically classify events based on a pre-existing classification system. Although information scientists devote considerable attention to assessing the performance of different supervised learning algorithms and feature representations, the questions asked are often less directly relevant to the more practical concerns of social scientists. The first question prospective social science users are likely to ask is, How well do such methods work? The second is, How much human labeling effort is required? The third is, How do we assess whether virgin cases have been automatically classified with sufficient accuracy? We address these questions in the context of a particular dataset—the Congressional Bills Project—which includes more than 400,000 bill titles that humans have classified into 20 policy topics. This corpus offers an unusual opportunity to assess the performance of different algorithms, the impact of sample size, and the benefits of ensemble learning as a means for estimating classification accuracy.