A discriminative and semantic feature selection method for text categorization

Text categorization is an important and critical task in the current era of high volume data storage and handling. Feature selection is obviously one of the most important steps in text categorization. Traditional feature selection methods tend to only consider the correlation between features and categories, and have in the main ignored the semantic similarity between features and documents. To further explore this issue, this paper proposes a novel feature selection method that first selects features in documents with discriminative power and then computes the semantic similarity between features and documents. The proposed feature selection method is tested using a support vector machine (SVM) classifier upon two published datasets, viz. Reuters-21578 and 20-Newsgroups. The experimental results show that the proposed feature selection method generally outperforms the traditional feature selection methods for text categorization for both published datasets.