Federated learning for breaking data silos in smart governance: a privacy-preserving framework for cross-agency collaboration

The realization of smart governance highly relies on the effective integration and collaborative utilization of cross-departmental government data, yet data silos that have formed over time and the privacy compliance risks faced by traditional centralized sharing models severely constrain the improvement of collaboration effectiveness. Addressing this challenge, this study proposes the FedGov privacy-preserving federated learning framework for smart governance scenarios, designing a three-layer system architecture comprising data, computation, and coordination layers to support multi-departmental heterogeneous data collaboration, and developing the FedGov-DP algorithm integrating dual mechanisms of differential privacy and secure aggregation to realize the "data usable but invisible" cross-departmental collaboration paradigm. Systematic experiments simulating government scenarios based on public datasets demonstrate that the proposed framework effectively breaks down data silos and achieves significant collaboration gains, the differential privacy mechanism effectively defends against membership inference attacks, and the method exhibits good adaptability to moderate data heterogeneity common in government scenarios. This study extends the application boundaries of federated learning in the public governance domain, provides a new technical pathway for addressing the government data silo dilemma, and the constructed framework with parameter configuration guidelines offers technical support and practical reference for smart governance digital transformation.