A Knowledge Graph-Based Framework for Identifying Audit Risks in Multinational Corporations

The globalization of markets and the increasing complexity of multinational corporations (MNCs) have posed significant challenges to modern audit risk identification. Traditional methods, often reliant on manual processes and static rule-based systems, struggle to address the dynamic and nonlinear interdependencies of risk factors inherent in MNC operations. In this study, we propose an innovative approach combining knowledge graphs (KGs) and machine learning (ML) techniques to overcome these limitations. KGs provide a structured representation of entities and their relationships, enabling the synthesis of complex, interconnected data. By integrating graph-based features with advanced ML models, such as XGBoost and deep neural networks, we uncover latent patterns and predict audit risks with high precision. A domainspecific knowledge graph is constructed using structured and unstructured data, including financial statements, compliance reports, and regulatory filings. Graph embeddings, generated through algorithms like Node2Vec and GraphSAGE, serve as input features for risk prediction models. Empirical experiments on a dataset of $\mathbf{5 0, 0 0 0}$ samples from $\mathbf{2 0 0}$ MNCs across 10 industries demonstrate the model’s superior performance, achieving an accuracy of $92.5 \%$, an $F 1$-score of $\mathbf{9 0. 8 \%}$, and an AUC of $\mathbf{9 4. 3 \%}$. The results highlight the utility of integrating KGs with ML for audit applications, offering enhanced scalability and accuracy in identifying high-risk scenarios. This research contributes to the development of intelligent auditing frameworks, addressing the growing demand for automated, data-driven risk assessment in complex global financial systems.