JD Digits' Intelligent Anti‑Fraud Platform: AI‑Driven Real‑Time Fraud Detection and Knowledge‑Graph Solutions

The platform integrates machine learning, big data, real‑time and graph computing to build a comprehensive intelligent anti‑fraud system that addresses marketing acquisition, ad placement, account, payment and asset security across more than 600 business scenarios. A high‑availability AI cluster handles over 1 trillion complex calculations daily with millisecond response times, proving its robustness during major sales events such as 618 and Double‑11.

Its core includes an auto‑adversarial machine learning engine that uses small‑sample learning and graph neural networks to automatically generate features, select models, recommend strategies and counteract fraud in real time, shifting defense from passive to proactive.

The team introduced an unsupervised heterogeneous graph neural network embedding model (HDGI) that preserves both node attributes and graph structure, enabling minute‑level detection of fraud groups among billions of nodes and edges with over 99% accuracy.

To address sparse or missing training data, a small‑sample knowledge‑graph completion algorithm predicts relationships between entities, dramatically reducing model development time from days to minutes and providing unattended real‑time security.

AI technologies such as a multi‑modal face‑liveness detection algorithm have passed national financial security certifications, protecting user accounts against sophisticated attacks.

In the past year, the team published more than 20 AI research papers at top conferences (NeurIPS, ICML, KDD, AAAI, etc.), including innovations like the Robust Conditional Generative Adversarial Network (RCGAN) and fast data‑screening for robust SVMs, which support rapid model iteration in fraud scenarios.

Overall, the platform demonstrates how AI, big data and graph analytics can evolve anti‑fraud from manual rule‑based methods to autonomous, self‑learning protection for digital finance.