Data Authorization and Federated Learning Execution Method Driven by Smart Legal Contracts

In the era of the digital economy, data has emerged as a core production factor for social innovation. Federated Learning (FL) enables collaborative model training while keeping data localized (characterized by “moving models instead of data”). This shifts data circulation from “centralized data sharing” to “collaborative model computation.” However, current FL architectures lack effective mechanisms for data rights confirmation and benefit allocation, as well as the legal responsibility definition for data authorization and FL execution. To address this issue, this paper proposes a smart legal contract-driven approach for data authorization and execution in FL. We design a federated learning governance framework based on the SPESC language for smart legal contracts, which facilitates the publication, assignment, and monitoring of federated learning tasks through contractual clauses. Within this framework, an Authorization and Execution Management Platform (AEMP) is designed to employs its Data Authorization Module for implementing a cyclical “offer–acceptance–execution–arbitration” process via standardized contract templates. By integrating decentralized identifiers (DIDs) and blockchain technology, the platform ensures identity authentication of contracting parties and enforces data authorization through self-executing contract clauses. Breach and arbitration clauses are also incorporated to supervise data ownership confirmation and rights allocation, ensuring compliance in data usage and operational rights among local training nodes and central model aggregation node. Furthermore, the federated computation module utilizes contract templates to configure computing tasks within the federated system and to oversee the responsibilities and accountability of participants during execution. Experimental evaluations demonstrate the feasibility of automated execution and on-chain traceability of data authorization clauses, ensuring identity compliance and transparency in federated learning. Additionally, a federated computing contract template is also proposed for analyzing the performance of node selection algorithms. The results indicate that the model training process under this framework is stable and exhibits a fast convergence rate. This research offers a novel approach to advancing the digital transformation of the data factor market.