Research on the Construction of Enterprise Compliance Management System in the Context of Deep Learning

Migration learning is a kind of deep learning that plays an important role in risk prediction and warning. In this paper, we use the Triplet-loss representation learning technique to map data samples of the same category to adjacent spatial regions, map data samples of different categories to different spatial regions, and add domain adaptive computation to complete domain adaptation. XGBoost performance is also optimized to improve the prediction results. After completing the risk prediction model construction, this paper also incorporates indicator weights to propose a risk warning model based on a weighted knowledge graph. The analysis found that the model accuracy of the compliance risk prediction model is 1.155% higher than that of two traditional prediction models on average, and the indicators of accuracy, F1-score, and AUC are all over 0.99, which indicates that it has high accuracy. After applying the optimized risk management system in an enterprise, it is found that the weight of the indicators of the enterprise’s compliance management system reaches 25.71%, the combined weight of the legal audit system and the risk prevention and control mechanism is 16.23% and 14.10%, respectively, and the combined weight of the two indicators of the appraisal and evaluation of the performance of the compliance management and the accountability for the violation of the law exceeds 8%. The weights of the indicators related to the degree of business standardization are close to 16.51%, indicating that the enterprise’s daily production and operation activities are standardized and orderly, and the perfection of the compliance management system is in line with the real situation of the enterprise. It can be seen that the risk management system makes accurate judgments on the compliance risk of the selected enterprise, which reflects the feasibility of the application of the new model. © 2024 Chunguang Li, published by Sciendo.