A machine learning based improved native Bayes algorithm proposed to solve the text classification problem in automatic document review field.Firstly,it improves naive Bayes algorithm and applies it as the classifier.Then a genetic algorithm is adopted to train all the feature weights.Finally,a table and figure position based identification algorithm is used to improve the results.The experimental results show that the algorithm performs better than traditional (K-nearest neighbors) KNN and naive Bayes in most cases,especially when the sample sets have more wrong samples.It can improve the accuracy of automatic document review effectively.