Although existing implemented regulatory responses have been effective in reducing the incidence of financial fraud, the regulatory measures are still not comprehensive enough. In this paper, Benford’s law is used as the entry point of the research, and the collected conducted financial data are preprocessed to circumvent the influence of interfering information on the research results. With the theoretical support of relevant definitions, the financial selection algorithm based on conditional dynamic mutual information is designed, and the idea and process of the algorithm are described in detail. On this basis, the variational autoencoder and one-dimensional graph convolutional neural network are used to construct a financial data fraud recognition model, and the model is analyzed by example verification. After analyzing and calculating, it can be concluded that the precision, recall, and F-Score of SVM are 0.8923, 0.8517, and 0.8715, and the values of RF indicators are 0.7672, 0.7314, and 0.7489, while the values of this paper’s method are 0.9728, 0.8896, and 0.9293, so the OCGVAE algorithm financial data fraud recognition process has the highest priority. The research in this paper is conducive to the good development of data quality in the financial market, effectively protect the rights and interests of investors and reduce the investment risk caused by information asymmetry.
