For the task of financial risk identification, this paper takes the financial text information of the company as the data source, and on the basis of the hierarchical attention network HAN, uses the long text classification model XLNet _ HAN that combines the double-level attention mechanism and the generalized autoregressive pre-training language model XLNet to carry out financial fraud identification and detection. Considering that the research object of this paper is the classification task of extremely unbalanced text datasets, this study adds a class center measurement layer between the coding layer and the loss function layer to further improve the accuracy of classification. The experimental results show that the financial risk identification accuracy of the model proposed in this paper is at least 70%. It shows that big data technology can effectively identify risk factors in financial risk identification, thereby providing reliable financial management reference for enterprises.