Federated Learning-Based Legal Compliance Detection of Financial Information in Private Educational Institutions
Danyang Liu, Wenzhou Shu, Leru Sun, Ziqing Huang, Zhen Tian, Limin Wei, Peng Leng · 2025
To address the challenges of compliance detection under highly heterogeneous financial data and rapidly evolving regulations in private educational institutions, this paper proposes FedLC—a federated learning-based legal compliance detection framework. FedLC introduces a cross-institution adversarial feature alignment mechanism, integrates a dynamic legal knowledge graph with graph neural embeddings to enhance regulatory adaptability, and incorporates dual-gradient privacy optimization along with a risk feedback heatmap for interpretable modeling. Experiments conducted on real financial data from multiple educational institutions demonstrate that FedLC outperforms mainstream federated models in terms of F1-score, error rate, and communication cost, highlighting its strong practical applicability for deployment.