The study uses Fudan Chinese text and Sogou Chinese document as the research object. It improves the Chinese text classification accuracy and recall rate. And it analyzes and obtains the best contribution value of the feature words. Based on naive Bayes classification method, improved TFIDF keyword extraction and weight calculation, the TNBIF model classification method is proposed and implemented on the Spark platform. The experimental results show that the Chinese text classification is applied by the TNBIF model. The accuracy is as high as 95.49%, which is 5.41% higher than the traditional text classification method and the recall rate is increased by 6.64%. This study obtains an optimal contribution of 0.95.