Text feature selection is a key issue in natural language processing. Due to the high-dimensional and sparsity of text features, based on the filter feature selection algorithm TF-IDF, the genetic algorithm was used to optimize the text features. To maximize the fit of the subsequent text classification algorithm, while not effecting the accuracy of the text classification, reduce the feature dimension to reduce the prediction time. Experiments show that compared with a single filtered text feature selection algorithm, the algorithm can effectively reduce the number of selected text features (reduce the feature dimension) and effectively improve the text classification ability.