Aerospike based NoSQL database has been leveraged for data analytical and Fraud detection use cases at PayPal Risk for ~3+ years. With existing Petabytes scale data storage support and more than 99.99+% availability from Aerospike Database Platform, a New Online Graph linking use case was introduced for real-time linking and fuzzy linking data analytical features on AI models.
To support powerful graph (fuzzy) linking cases for better fraud mitigation, a solution of highly-scalable graph database based on Aerospike/Gremlin graph query language with 50 billion+ vertices/edges will be introduced in this session for both online graph linking cases with high performance and offline graph analytical solution with high throughput requirements.
PayPal 风险管理部门通过应用基于 data 和 AI 的解决方案检测 PayPal 平台的欺诈交易。从三年前开始 Aerospike 做为主要的 NoSQL 技术被用以快速存取风险管理场景所需要的各种 KV 数据,到目前为止已经有 20+ Aerospike 集群、PB 以上的数据,而且可用性达到了 4 个 9 以上。基于 Aerospike 的良好的性能和稳定性,我们构建了实时的图的连接以及模糊连接查询和计算,用以支持风险管理系统对图连接数据类型的需要。
目前整个实时图计算平台基于 Gremlin 接口抽象,后端存储主要是 Aerospike 但不限于单一的存储解决方案,这有利于在其它不同业务要求下的系统扩展。在其上我们构建了 500 亿以上的点和边的风险管理连接图且可以支持动态(模糊)连接数据的扩展。通过使用异步化、批量化、缓存等优化方法满足实时风险管理在高维多跳的图连接查询和运算的低延迟高吞吐的需求。
听众受益
1. How Aerospike NoSQL DB has been leveraged at PayPal Risk Fraud Detection Cases;
PayPal 在欺诈检测场景如何利用 Aerospike 支持 PB 级别实时数据管理;
2. How to build 50 billion + online linking graph with high query performance latency using Gremlin;
如果基于 Gremlin 和 NoSQL(Aerospike)支持 500 亿以上的点和边的实时图查询和运算;
3. How graph linking based data analytical solution, AI models are used for solution in PayPal Risk team.
PayPal 如何利用实时的图连接运算加强基于数据和 AI 的风险管理能力。