Bank of Communications (BoCom)
Galaxybase Enpowered Real-Time Credit Card Application Fraud Dection System
Customer Profile
Bank of Communications (BOCOM), founded in 1908, is one of the oldest and most influential commercial banks in China. As a Fortune 500 state-owned commercial bank, it offers a comprehensive range of financial services, across more than 3000 domestic and global branches.
Business Challenges
Core Anti-Fraud Challenges for Credit Card Centers: Data Scarcity and Outdated Rules
For credit card centers, anti-fraud initiatives face unique challenges amid new economic policies and stringent data regulations: a persistent shortage of representative fraud samples due to limited data access and evolving fraud patterns, resulting in suboptimal machine learning model performance; rigid, slow-to-update static rule systems that fail to keep pace with increasingly sophisticated and adaptive fraudulent tactics in digital era; and an over-reliance on manual review processes that are not only operationally inefficient but also highly costly—especially as transaction volumes surge
solution
BoCom’s Credit Card Center built a real-time anti-fraud platform using Galaxybase Graph Database. By integrating expert rules, machine learning algorithms, and graph-based network metrics, the platform automated and digitized the identification of individual credit risks and gang fraud risks—enabling end-to-end real-time risk control for credit card applications.
Our Approach
Constructing Application Graph
By integrating multi-line internal data—encompassing applications, credit reviews, and collections—with external credit data, we have constructed a comprehensive 360-degree user panoramic application graph designed to identify fraudulent rings and track risk propagations. This sophisticated graph comprises 980 million nodes, 3.66 billion edges, and 43.5 billion properties (as of 2022). These elements are derived from user profiles, transaction records, contact information, and other associated features related to fraud syndicates.
Developing Graph-based Risk Metrics
With the support of the panoramic application graph, the platform employs a hybrid methodology that integrates rule-based models, graph algorithms, and adaptive graph feature learning to identify gang fraud and its underlying network patterns. It then computes a comprehensive risk score for each application based on its structural position and behavioral associations within the graph. These graph-derived risk labels are subsequently integrated as predictive features within downstream machine learning models.
Real-time Fraud Detection
When a new application is submitted, the platform extracts entities and relationships pertaining to the applicant from the application data and dynamically inserts them into the application graph stored in an OLTP in real-time. Within milliseconds, dozens of multi-hop deep-link graph metrics are computed. The applicant’s risk scores are derived from these metrics, and the risk properties of associated entities are updated accordingly.
These results are immediately sent to BoCom’s downstream credit risk detection system for a comprehensive assessment of the applicant’s risk profile. Simultaneously, the same data is synchronized into an OLAP cluster for large-scale graph pattern analysis aimed at identifying fraudulent networks and behaviors. The insights generated are fed back to the OLTP cluster in a controlled, near-real-time manner. ensuring that the system remains up-to-date and responsive to emerging risks.
Business Value
Addressing the limitations in machine learning caused by insufficiently-labeled fraud samples and the rigidity of traditional expert-based rules, this end-to-end solution supports real-time risk control throughout the credit card application and underwriting process. It has enabled the interception of a significant volume of fraudulent applications, and resulted in a remarkable 50% reduction in the post-issuance fraud rate, significantly strengthening BoCom’s defense against fraudulent activities.
Why Choose Us
High-performance graph query and computation
The system computes tens of graph features within milliseconds, significantly enhancing risk control efficiency and customer experience.
Outstanding detection of sophisticated fraud patterns
The system effectively identifies fraudulent applications during the pre-issuance phase, achieving a fraud detection precision rate of 99.26% and a recall rate of 52%. This capability enables the detection of hidden fraud that is difficult to uncover through traditional methods.
Improved Credit Review Efficiency
By automating previously manual and multi-system review processes into near-instantaneous automated analysis, the system reduces manual effort by 90% and improves operational efficiency by nearly 1,000 times.
Create Link is the first domestic commercial graph database supplier with fully independent intellectual property rights, dedicated to providing world-class graph database products and graph intelligence services.
Business Consulting
400-882-6897
Pre-sales Consulting
0571-88013575、0571-88016275
Enterprise Email
partner@chuanglintech.com
Media Cooperation
Marketing@chuanglintech.com
Address
Room 605, Building 6, No.666, Zhenhua Road, Sandun Town, Xihu District, Hangzhou, Zhejiang, China
