earticle

영어

The text representation, “bag of words” or vector space model, is widely used by most of the classifiers in text categorization. All the documents fed into the classifier are represented as a vector in the vector space, which consists of all the terms extracted from training set. Due to the characteristics of high dimensionality, feature selection algorithm is usually used to reduce the dimensionality of the vector space. Through feature selection, each document is represented by some representative terms extracted from the training set. Although the classification results based on this document representation methodare better, it is inevitable that some documents may contain few even none representative terms, and these documents must be misclassified. In this paper, we proposed a new text representation method, KT-of-DOC, which represents one document using some key terms extracted from this document. We selected key terms of each document based on six feature selection algorithms, Improved Gini Index (GINI), Information Gain (IG), Mutual Information (MI), Odds Ratio (OR), Ambiguity Measure (AM) and DIA association factor (DIA), respectively, and evaluated the performance of two classifiers, Support Vector Machines (SVM) and K-Nearest Neighbors (KNN), on three benchmark collections, 20-Newsgroups, Reuters-21578 and WebKB. The results show that the proposed representation method can significantly improve the performance of classifier.