Abstract:This paper proposed a method to analyzing subjective text. The method uses various strategies to stand for text with feature vectors, and uses SVM to classify text according to the property of subjectivity and objectivity after eliminating the rundant and irrelevant features using feature selection algorithm. The feature selection algorithm in the paper bases on SIMBA. We improve the original SIMBA on the way of iteration and the measure of similarity through experiment, and overcome the instability when putting into application. In the experiment done on English and Chinese corpus respectively, the accuracy overperforms that by SVM algorithm alone and the F-MEASURE is better than that by the baseline method on same corpus.