Best-match method used in co-training algorithm

Abstract

Since 1998 there has been significant interest in supervised learning algorithms that combine labeled and unlabeled data for text learning tasks. The co-training algorithm applied to datasets which have a natural separation of their features into two disjoint sets. In this paper, we demonstrate that when learning from labeled and unlabeled data using co-training algorithm, selecting those document examples first which have two parts of best matching features can obtain a good performance. © Springer-Verlag Berlin Heidelberg 2007.