Bank of Communications
Building a real-time online credit card application anti-fraud system using the Galaxybase graph database
Customer Profile
ank of Communications, founded in 1908 (34th year of the Guangxu Emperor reign), was restructured domestically in 1986 with its headquarters in Shanghai. It is China's first nationwide joint-stock commercial bank. As one of China's major financial service providers, Bank of Communications' business scope includes commercial banking, securities, trusts, financial leasing, fund management, insurance, offshore financial services, and more.
In 2004, Bank of Communications introduced advanced technology and management from HSBC to establish the Pacific Credit Card Center, creating credit card products with global and local advantages. Over the years, the bank's credit cards have been at the forefront of business innovation.
In 2022, the issued card volume reached 74.5 million, ranking among the top ten in China.
Business Challenges
In the new economic environment, Bank of Communications' credit card center faces high delinquency rates.
In 2019, the bank's credit card non-performing loan ratio reached 2.20%. On one hand, existing static expert rules are ineffective against dynamic and evolving fraud methods. On the other hand, the concealment of fraud syndicates results in insufficient fraudulent samples, leading to traditional machine learning methods having poor performance in effectively detecting large-scale and covert syndicated fraud.
Manual verification for credit card fraud is inefficient and costly.
Post-verification of suspicious credit cards relies on manual intervention, which is costly and time-consuming. The long training cycle for skilled reviewers leads to high overall costs.
solution
Bank of Communications' credit card center proposed building the Athena Cognitive Graph Platform based on the Galaxybase graph database. This platform aims to address the large volumes of structured, semi-structured, and unstructured data within the card center by utilizing techniques such as entity extraction, knowledge fusion, and knowledge processing to construct a large-scale cognitive graph encompassing industry knowledge, common sense knowledge, and cognitive knowledge.
Specific programs
Graph Construction for Applications
By integrating internal application data, credit reviews, collections, and external credit information, leveraging the associated features of fraud syndicates, entities from application data are represented as nodes, and relationships between applicants/applications and their information are depicted as edges. This process results in constructing a 360-degree panoramic application graph from user, transaction data, contact details, etc., with a scale of 9.8 billion nodes, 36.6 billion edges, and 435 billion attributes.
Development of Graph-based Risk Indicators
Utilizing the comprehensive application graph, in conjunction with rule models and graph algorithms, adaptive graph feature learning is employed to uncover syndicated fraud and its associated features. This process involves supplementing risk labels for historical application entities and enhancing risk control dimensions
Real-time Fraud Detection
New incoming application entries are dynamically incorporated into the graph in real-time, enabling the real-time computation of risk associated with multi-hop deep network relationships. This process integrates with existing control systems to conduct fraud identification in the pre-loan application stage, achieving real-time detection and prevention of application fraud risks. This approach addresses deficiencies such as insufficient black samples and inflexible expert rules, automating the identification of large-scale and covert fraud
Value of Use
The implementation of a real-time, comprehensive, full-lifecycle credit card user management, and risk control system has resulted in a significant decrease in Bank of Communications' credit card delinquency rate, demonstrating notable effectiveness.
High-performance online query calculations, millisecond response times, real-time risk prevention and control
New application data is dynamically integrated into the graph in real-time, enabling automatic real-time fraud detection for each application. Complex deep-link graph indicators can be computed within one second, shifting from post-event analysis to proactive intervention, reducing systemic and large-scale syndicated fraud risks.
Accurate identification and uncovering of hidden clues
Fraud risk patterns are explored and identified among historical application data to reduce misjudgment risks. With billions of applications, post-issuance fraud detection rates reach 99%, with a recall rate of 52%.
Integration of cross-source data enhances credit review efficiency
Manual reviews across systems that took tens of minutes are now automatically completed at the millisecond level, resulting in a 90% reduction in labor costs and a thousandfold increase in efficiency.
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
L12, Building E2, Zhi Hui Zhong Chuang Center, No.3 Xiyuan No.8 Road, Xihu District, Hangzhou, China
Contact Us