-
英文自我介绍
-
根据简历问问题,主要问的是数据实习
-
涉及到的技术问答(都是在谈业务题的时候顺带出来的)
○ python:python有哪些数据结构、list/dict/tuple/dataframe的区别、用什么包 ○ 模型:各种指标(准确率、查全率、查准率),Roc曲线、AUC
-
问答题:
-
给一张打印好的表,上面字段分别是卖家id、买家id、交易日期、交易金额、识别出欺诈的日期、欺诈金额(如果识别出欺诈,则欺诈金额为交易金额)
回答五个问题:`` ○ sql:七、八、九月份累计交易金额量 ○ sql:七、八、九月份最后一天观测到的累计交易金额量(这个题记不清了,但是重点在于识别出欺诈的日期) ○ 写出上一题的伪代码 ○ 你觉得交易的风险增加了吗(表面上看欺诈笔数是逐月递减的,但是因为识别出欺诈的日期通常在交易后的许多天甚至几个月后,所以不好判断?) ○ 你从表中还可以看出什么:有一个卖家id出现许多欺诈交易(我只答了这一个,可能还有别的),然后会引出业务题
-
有两张表:
求出每个国家发生交易的总额(买家和卖家在两个国家的分别计入):select country,sum(amount) as sum from ((select t.*,u1.country as country from transaction as t left join user as u1 on t.seller_id=u1.user_id) union (select t.*,u2.country as country from transaction as t left join user as u2 on t.seller_id=u2.user_id))new group by country
-
有一张表:
求出每个userid连续交易的最长天数(适用于一天有多笔交易的情况)select userid,max(cons_num) as max_cons_num from (select userid,cons_no,count(cons_no) as cons_num from (select *, datediff(day,'2019-01-01',dateid) as day, row_number() over(partition by userid order by userid,dateid) as rownumber, datediff(day,'2019-01-01',dateid)-row_number() over(partition by userid order by userid,dateid) as cons_no from (select * from trans group by userid,dateid)a)b group by userid,cons_no)c group by userid order by userid
求出每个userid连续不交易的最长天数:
select userid,max(uncons_day) as max_uncons_day from (select *, min(dateid) over (partition by userid order by userid,dateid rows between 1 preceding and 1 preceding) as preceding, datediff(day,min(dateid) over (partition by userid order by userid,dateid rows between 1 preceding and 1 preceding),dateid) as uncons_day from (select * from trans group by userid,dateid)a)b group by userid
-
业务题
卖家/买家欺诈特征 卖家/买家有哪些欺诈手段 怎么处理 如果你是卖家/买家怎么应对这些处理 怎么分层(记得先问她分层的目的是什么) 从什么维度分层 怎么合伙从paypal骗钱 注:以上问题他们会深挖,取决于你上述问题回答了什么