大数据风控实践 唐会军

—1—大数据风控实践GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC2016.07垃圾文本评论、弹幕，私信…行为欺诈刷榜、刷人气…信贷欺诈薅羊毛电商，O2O…应用欺诈无处不在GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC反欺诈行业面临挑战01对手越来越专业，变化越来越快02反欺诈团队人手紧，疲于应付03依赖更多维度数据消除不确定性GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC反欺诈核心GITCGITCGITCGITC数GIT据CGITC+GITC策GIT略CGITCGITCGITCGITC全方位数据基础数据（账号，手机号，IP，设备，域名，邮箱）GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC全方位数据行为数据（注册，登录，发帖，浏览，点赞，购买）GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC交叉验证•横向：多渠道，多业务•纵向：时间维度GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC时效性•每项数据有时间戳和有效时间窗口•数据过期前用各种数据源更新GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC•过期数据有效权重下降重复•身份标识重复：账号，IP，手机，设备•行为重复：注册，发帖，刷榜GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC•内容重复：文本，联系方式隐私GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC策略GITCGITCGITCGITC数GIT据CGITC+GITC策GIT略CGITCGITCGITCGITC规则引擎普通规则模型规则自定义函数GITC调G度I优TC先G级ITCGITCGITCGITCGITCGITCGITCGITCGITCGITC模型专业化联系方式引擎敏感词引擎画像引擎色情引擎广告引擎重复度引擎上下文引擎GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC规则实验镜像流量测试对比评估上线GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC开源工具在线离线GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC开源工具升级后，单台服务器吞吐（QPS）提升将近10倍GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC合作共享GITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITCGITC—18—谢谢[e-mail]