Large-Scale Graph Platform Dxm Eros for Financial Risk Control

The talk is organized into four parts: background of fintech evolution, the functions of the Dxm Eros platform, its applications, and a summary with future outlook.

1. Fintech development stages – From pre‑1986 informationization, through the 1987‑2009 internet finance boom, to the current deep integration era where big data, cloud, AI, and blockchain reshape financial services.

2. Evolution of financial risk control – Describes the shift from manual review to rule‑based systems, then to machine‑learning models, and finally to knowledge‑graph approaches that capture relationships among entities.

3. Graph in financial risk control – Traditional relational databases cannot efficiently model complex relationships; graph databases store billions of nodes and edges, enabling real‑time risk detection and fraud identification.

4. Dxm Eros platform

• Infrastructure : massive heterogeneous data pool, compute clusters, high‑performance databases.

• Graph database : stores relational data such as person‑company‑document links.

• Platform capabilities : supports visualization, data fusion, auto‑modeling, and a library of classic and custom graph algorithms for credit‑pre, credit‑mid, and post‑loan scenarios.

• Graph storage module : handles trillions of nodes/edges with millisecond‑level query latency and integrated graph analytics.

• Graph analysis module : provides vertex, edge, and community analysis, supporting algorithms like GraphSAGE, GAT, and graph representation learning.

• Graph modeling module : AutoGraph automatically generates graph embeddings and supports incremental training on heterogeneous graphs.

• Visualization module : offers macro‑level graph layout, micro‑level node/edge details, community detection, and tools for fraud, AML, and audit workflows.

Applications

• Intelligent graph mining : end‑to‑end pipeline (data integration, model definition, training, tuning, scoring) enabling credit risk and fraud models with billions of vertices and edges.

• Intelligent anti‑fraud : visual and analytical tools to identify suspicious nodes, communities, and paths, supporting both individual and group fraud detection.

• Intelligent anti‑money‑laundering : tracks suspicious transaction flows and structures using graph‑based path analysis.

• Intelligent audit : combines OCR, NLP, and knowledge graphs for automated, rapid credit‑application processing and risk assessment.

Summary and outlook – Graph machine learning is widely applicable across the entire credit lifecycle; future work aims to lower the barrier to graph learning, enrich visualization, and provide industry‑wide solutions.