Training data cleaning for text classification

In text classification (TC) and other tasks involving supervised learning, labelled data may be scarce or expensive to obtain. Semi-supervised learning and active learning are two strategies whose aim is maximizing the effectiveness of the resulting classifiers while minimizing the required amount of training effort; both strategies have been actively investigated for TC in recent years. Much less research has been devoted to a third such strategy, training data cleaning (TDC), which consists in devising ranking functions that sort the original training examples in terms of how likely it is that the human annotator has misclassified them, thereby providing a convenient means for the human annotator to revise the training set so as to improve its quality. Working in the context of boosting-based learning methods we present three different techniques for performing TDC and, on two widely used TC benchmarks, evaluate them by their capability of spotting misclassified texts purposefully inserted in the training set.