AI-Driven Financial Knowledge Graphs: Bridging Traditional Finance and Blockchain Ecosystems with Graph Neural Networks

AI-Driven Financial Knowledge Graphs: Bridging Traditional Finance and Blockchain Ecosystems with Graph Neural Networks

연구 분야: Artificial Intelligence

학회: 2025 International Conference on Electrical, Computer and Communication Engineering (ECCE)

Integrating traditional financial systems with blockchain-based ecosystems represents a significant advancement in the evolution of modern financial infrastructures. This paper proposes an AI-driven methodology that employs Knowledge Graphs and Graph Neural Networks (GNNs) to bridge these two domains, facilitating seamless operations and enhanced decision-making. Using historical cryptocurrency data as a proxy for blockchain activity, we construct a unified knowledge graph that captures the intricate relationships between cryptocurrencies and traditional financial metrics. The proposed GNN model demonstrates superior performance across multiple tasks. For node classification, it achieves an F1-score of 88.9%, outperforming state-of-the-art models such as GraphSAGE (85.0%) and GAT (85.9%). In link prediction, the model achieves an AUC of 94.7%, surpassing Node2Vec (89.5%) and GraphSAGE (91.2%). The model also excels in anomaly detection, with an F1-score of 89.5%, significantly improving over baselines such as Autoencoder (82.5%) and GAT (86.3%). Additionally, scalability analysis shows that the model maintains reasonable training and inference times, even for large graphs with up to 500,000 nodes. The results confirm the approach’s effectiveness but highlight limitations like reliance on proxy data and high training costs for large graphs. Future work will explore the incorporation of real blockchain transactional data, automated feature engineering, and domain-specific applications such as decentralized finance.