The data division of large-scale text is a key problem in data governance, but the traditional Chinese document modeling method is easy to ignore the contextual semantic relationship and the hierarchical structure of the document. To solve the above problems, a text data governance method based on hierarchical characteristics and DPCNN is proposed. Firstly, the hierarchical feature information of text is extracted by BERT model. Then the vector combined with the full text information is passed into DPCNN model, after passing through the pyramid pooling layer; Finally, the data is divided through the full connection layer. This method can effectively improve the prediction accuracy of sparse feature text data.