A new approach for text feature selection based on OWA operator

Feature selection has a significant role in the precision of text classification algorithms. In this regard, various approaches exist such as information Gain, Chi Square, Document Frequency, Mutual Information, etc. To improve the classification effectiveness combination of some input features may help a lot. In this paper, a new approach based on Ordered-Weighted Averaging (OWA) is proposed for combining two feature set selection algorithms named Information Gain (IG) and Chi-Square(X2). The proposed approach is applied on the dataset of Reuters-21578. Obtained results show that OWA operator in general outperforms averaging and maximizing operators which have been used before in the text classification field. To evaluate the capability of OWA in comparison with averaging and maximizing operators, micro-averaged F1 and macro-averaged F1 measures are used.