A Simple and Fast Term Selection Procedure for Text Clustering

Text clustering is a theme that is receiving considerable attention nowadays in areas such as text mining and information retrieval. A starting point for clustering methods applied on unstructured document collection is the creation of a vector-space model usually known as bag-of-words model [1]. Documents are then usually described by a matrix which happens to be huge and extremely sparse which is due to the exceeding number of terms describing the set of documents. Although several techniques can be employed to reduce this number, the final figure is still high thus leading to a feature space of high dimensionality. This paper presents a simple procedure that not only considerably reduces the dimensionality of the feature space and hence the processing time, but also produces clustering performances comparable or even better when confronted with the full set of terms.