Abstract: For feature extraction and classification in text retrieval,a feature selection and sorting learning method based on embedded space support vector machine is proposed.Unlike combination methods commonly used in multi-classification feature selection,the proposed method can transform an ordered classification into a two-classification problem,then choose the most effective feature from the whole.At the same time,comparing with the existing Ranking SVM,the learning samples number of the proposed method just has a linear level increasing during the conversion process,and the retrieval speed is greatly improved.The experimental results on both artificial and standard data sets show that the proposed method can better solve the feature selection and sorting problem in text retrieval.