A semi-supervised cluster-and-label approach for utterance classification

In this paper, we propose a semi-supervised cluster-and-label algorithm for utterance classification. The approach assumes that the underlying class distribution is roughly captured through– fully unsupervised–clustering. Then, a minimum number of labeled examples is used to automatically label the extracted clusters so that the initial label set is ”augmented” to the whole clustered data. The optimum cluster labeling is achieved by means of the Hungarian algorithm, traditionally used to solve optimization assignment problems. Finally, the augmented labeled set is applied to train an SVM classifier. We compare this semi-supervised approach to a fully supervised version in which the initial labeled sets are directly used to train the SVMmodel.