Graph Computing for Risk Control in WeChat Pay: Platforms, Algorithms, and Practices

WeChat Pay, as a national‑scale payment tool, generates massive transaction data that attracts sophisticated fraud schemes. Traditional feature‑based risk control struggles with the sheer volume and cost of features, prompting a shift to a network‑centric perspective that treats users and fraudsters as interconnected graphs.

The new risk view emphasizes moving from individual feature engineering to global network engineering, recognizing that fraud often operates as organized groups rather than isolated actors. This insight drives the adoption of graph algorithms and graph databases for more effective detection.

We built a three‑horse‑carriage graph computing platform: a graph computation engine, a graph storage engine (graph database), and business‑specific algorithm design. Open‑source platforms such as Angel (a general big‑data platform) and Plato (Tencent’s internal graph engine) were co‑developed and deployed.

Speed is the primary criterion when selecting a graph platform; experiments showed that slower platforms dramatically increase model training time, hindering iterative improvement. High‑performance graph databases like EasyGraph (based on S2Graph) and TigerGraph enable sub‑second queries compared to minutes with traditional SQL.

Algorithmic innovations include gang identification, graph neural networks (GNN), propagation coloring (e.g., Personalized PageRank), and time‑series anomaly detection using HP filters and Ego‑TLSTM. Sample augmentation via look‑alike networks and connected‑component analysis further improve detection of low‑sample fraud patterns.

Practical applications span fraud detection, money‑laundering prevention, and device‑user network analysis. Converting heterogeneous graphs to homogeneous ones allows semi‑supervised GNN training, achieving AUC improvements when combined with traditional feature models.

The initiative exemplifies "technology for good": the team contributes to anti‑fraud, anti‑gambling, and anti‑money‑laundering efforts, receiving gratitude from affected users and reinforcing the mission to use advanced graph analytics for a fairer, safer digital ecosystem.