业务增益32NebulaGraph在携程风控场景下的落地实践-周伟诚携程风控团队

https://nebula-graph.com.cnhttps://nebula-graph.com.cn••••https://nebula-graph.com.cnhttps://nebula-graph.com.cn1https://nebula-graph.com.cn2NebulaGraphhttps://nebula-graph.com.cn欺诈识别率1阶邻居2阶邻居新客信息背景https://nebula-graph.com.cn2阶关联一年内业务需求https://nebula-graph.com.cnUIDcontactemailcardnorefidmobilephonechlorofpclientidvideventidrequesttimerequesttime图模型图谱建模原始数据格式https://nebula-graph.com.cnuid1mobile1clientid1 clientid2clientid3clientidNmobile1mobileN mobileN+1mobile2N mobile3NMobileN*N uid2热点导致关联数据巨大查询逻辑难点1https://nebula-graph.com.cn查询逻辑难点2uid1clientid1mobile1Uid2mobile2当笔交易中的介质元素尚未关联, 无法关联到下阶数据https://nebula-graph.com.cn解决查询逻辑uid1clientid1mobile1mobile2uid2mobile3uid3clientid3clientid2mobile2mobile3uid3uid5clientid5uid1clientid1mobile1client数量超过10个mobile2uid2mobile3uid2clientid3舍弃clientid2mobile2uid2clientid3mobile4mobile5uid4clientid4合并去重合并去重当笔订单元素一阶关联查询一阶关联结果整理合并二阶关联查询二阶关联结果整理合并https://nebula-graph.com.cn关键NGQL查询逻辑https://nebula-graph.com.cn项目成果关键指标◼◼原始信息1阶邻居40%2阶邻居15%实时图特征平台架构https