Text Classification Using Semi-supervised Clustering

In this paper, mixture models are used to classify documents. The basic assumption for the documents in a collection is that each class is composed of a number of mixture components. By identifying the components in the document collection, the classes of documents can thereby be identified from each other. A semi-supervised clustering method is proposed to identify the components (clusters), and further, unlabeled data is used to produce more accurate clusters in document collection to correspond the components of document classes. Experimental results show that the proposed method produces better performances than support vector machine (SVM) with linear kernel, and produces comparable performance with Bayesian classifier with expectation maximization (EM) in text classification.