Abstract: According to the feature of case text，this paper explores the special text preprocessing method and compares two effective feature selection methods.An improved model based on multi-variate Bernoulli model is proposed，due to the unbalanced distribution of criminal case categories.The experiment indicates that the improved Naive Bayes method performs better in the case text classification.

摘要： 针对案件文本的特点，提出了具有针对性的特殊文本预处理方法，并比较了两种有效的特征选择方法。针对案件类别分布不均衡的特点，提出了改进的多变量贝努里模型。实验结果表明，改进的多变量贝努里模型有效地提高了案件文本分类的准确率。