This article aims to improve the accuracy and speed of text classification. T * f algorithm is used to initially weigh the feature item, then stop words is used to shield specially meaningless words. Original probability distribution method and weighted L - E operator enable the features in the special positions or widely distributed to weight in exponential form, so that the better results converge faster. In this paper, by using the genetic algorithm, crossover operator and mutation operator, and adopting appropriate objective function, the retrieval process speeds up, and has a greater probability to get the optimal result. Hybrid algorithm is proposed, which can eliminate the synonyms and the characteristics of interference.