Distributed Clustering with Feature Selection for Text Documents Based on Ontology

Feature selection has been extensively used in supervised learning, such as text classification. It (Devaney and Ram 1997) minimizes the high dimensionality of the feature space and also offers improved data understanding which enhances the clustering result. The chosen feature set should consist of adequate data about the original data set. It is believed that feature selection can enhance the performance of text classification techniques by eliminating unnecessary and inappropriate terms from the corpus. Conventional feature selection techniques for classification are either supervised or unsupervised, depending on whether the class label data are needed for each document. Unsupervised feature selection approaches can be easily applied to clustering which uses document frequency and Term Strength (TS). But supervised approaches with the help of information gain and the 2 statistics can achieve better clustering performance than unsupervised techniques when the class labels of documents are available for the feature selection.