James Long 测试 Jev 从银行原始数据提取收款人描述
James Long 用 Jev 解决个人财务工具中最棘手的问题:从原始银行数据中生成良好的收款人描述,首次尝试即达到约 95% 的完成度,并指出还有多种优化提问方式。
作者试用新的 Jev 动作选择模型,认为其不太可能取代支付欺诈检测,因为许多特征为表格数据,表现不及 LightGBM,但考虑到该场景可能并非其目标用例,结果已令人印象深刻。
Time to start playing with #jev the new action selection model. Will it replace Fraud detection for payment. Well unlikely because a lot of the features are tabular. Underperforming a Lightgbm but it it already impressive given that this probably not the use case for it.