Co-Training on Textual Documents with a Single Natural Feature Set.
Jason Wei Siang Chan, Irena Koprinska, Josiah Poon · 2004
Abstract Co-training is a semi-supervised technique that allows classifiers to learn with fewer labelled documents by taking advantage of the more abundant unclassified documents. However, conventional co-training requires the dataset to be described by two disjoint and natural feature sets that are redundantly sufficient. In many practical situations datasets have a single set of features and it is not obvious how to split it into two. This paper investigates the performance of co-training with only one natural feature set in two applications: Web page classification and email filtering.