Existing governance in decentralized financial networks relies on manual review by human committees, which introduces subjective bias, creates a scalability bottleneck under high transaction volume, and fails to capture complex behavioural patterns and evolving network dynamics. To overcome these issues, we propose DeFiRank, a federated learning‑based pairwise ranking framework where multiple institutions collaboratively train a ranking model without sharing private transaction data. The model learns network‑wide behavioural signals such as counterpart diversity, cross‑application usage, and transaction frequency, all of which are indicative of strong network participation. The system ranks providers by the authenticity of their actions, promoting genuine contributors and demoting suspicious actors. Evaluation on a real blockchain transaction dataset shows the federated approach achieves NDCG@15≈0.87, modestly below the centralized baseline≈0.92, while matching its ranking stability (Kendall τ≈0.70). It enables reliable governance decisions, promoting 648 unique providers and demoting 39, without requiring centralized access to sensitive data.
Review