DeFiRank Uses Federated Pairwise Ranking for DeFi Governance

Summary

DeFiRank is a federated learning-based pairwise ranking framework for governance in decentralized financial networks. It addresses manual committee review, which the authors say introduces subjective bias, limits scalability at high transaction volume, and misses complex or changing behavioral patterns. Multiple institutions train the ranking model collaboratively without sharing their private transaction data. The model evaluates signals associated with genuine network participation, including counterparty diversity, use of multiple applications, and transaction frequency, then ranks providers by the authenticity of their activity to support promotion or demotion. On a real-world blockchain transaction dataset, the federated system reached an NDCG@15 of approximately 0.87, compared with approximately 0.92 for a centralized baseline. Its ranking stability matched the centralized setting, with Kendall's tau of approximately 0.70 in both cases. The system produced governance decisions involving 648 unique provider promotions and 39 unique provider demotions while avoiding centralized access to sensitive transaction data.