Traditional multi-attribute group decision-making (MAGDM) methods primarily rely on expert knowledge-driven information, often overlooking the value of objective data in decision-making processes. To address this gap, this paper proposes a novel dual-driven MAGDM method that incorporates knowledge-driven information, expressed through single-valued neutrosophic credibility numbers (SvNCNs), and data-driven information, represented by exact numbers (ENs). The primary innovations of this method include the development of the variable extended power geometric (VEPG) operator, which effectively aggregates knowledge-driven information from SvNCNs. The SvNCN Einstein variable extended power geometric (SvNCNEVEPG) operator is also introduced, and its mathematical properties are rigorously proven, offering an advanced approach to handling extreme values. To resolve ambiguity and uncertainty in decision analysis, a new subjective weight determination method, SvNCN-PIPRECIA, is introduced, complemented by an objective entropy-based weighting method. These are integrated into a new combined weight determination model using the Uninorm operator, enhancing the accuracy and reliability of the decision-making process. The dual-driven MAGDM method, combining knowledge-driven and data-driven information, improves decision-making comprehensiveness and precision through the dual-driven SPA-Entropy-PIPRECIA-MARCOS approach. The proposed methodology is validated through a case study on evaluating data trading platforms (DTPs), where sensitivity analysis of parameters and a comparative study with existing methods demonstrate the flexibility and scientific robustness of the approach.

In data analysis, privacy preserving is receiving more and more attention, privacy concerns results in the formation of “data silos”. Federated learning can accomplish data integrated analysis while protecting data privacy, it is currently an effective way to break the “data silo” dilemma. In this paper, we build a federated learning framework based on differential privacy. First, for each local dataset, the summary statistics of the parameter estimates and the maximum norm of the coefficient vector for the polynomial function used to approximate individual log‐likelihood function are computed and transmitted to the trust center. Second, at the trust center, gaussian noise is added to the coefficients of the polynomial function which approximates the full log‐likelihood function, and the parameter estimates under privacy is obtained from the noise/privacy objective function, and the estimator satisfies ‐DP. In addition, theoretical guarantees are provided for the privacy guarantees and statistical utility of the proposed method. Finally, we verify the utility of the method using numerical simulations and apply our method in the study of salary impact factors.