Semi-supervised clustering with soft labels

This paper devises a semi-supervised learning algorithm to cluster text documents. The proposed algorithm clusters multi-dimensional documents using the k-means algorithm. It initially reduces the dimensionality of the text so that the clustering algorithm can perform well in the low dimensional feature space. It also removes the irrelevant, redundant and noisy features from the corpus which may otherwise mislead the underlying algorithm. The proposed method employs pLSA algorithm to generate soft labels from these reduced feature subset and these labels along with the class labels guide the k-means algorithm. Experiments were conducted on Reuters-21,578 dataset and the results obtained showed that the proposed method outperforms many previous clustering algorithms without supervision.