Active Learning with Committees in Text Categorization: Preliminary Results in Comparing Winnow and Perceptron

Poster Spotlights

The availability of vast amounts of information on the World Wide Web has created a big demand for automatic tools to organize and index that information. Unfortunately, the paradigm of supervised machine learning is ill-suited to this task, as it assumes that the training examples are classified by a teacher -- usually a human. In this paper, we describe an active learning method based on Query by Committee (QBC) that reduces the number of labeled training examples (text documents) required for learning by 1-2 orders of magnitude.