sql - 从客户事件、SQL 中查找放弃的搜索

Question

在交易平台中，用户有三种不同的操作:LOGIN , SEARCH , ORDER .

被遗弃SEARCH操作定义为:客户LOGIN然后做一些SEARCH并且不要ORDER之前 下一个 LOGIN 。 现在我有一张 table :action_table ，录音customer_id , action , request_time .

我想知道如何通过 SQL 查找所有废弃的 SEARCH 操作？

---

更新: 这是一个简短的示例:

CREATE TABLE action_table(
   customer_id  VARCHAR(1) NOT NULL
  ,action       VARCHAR(6) NOT NULL
  ,request_time DATE  NOT NULL
);
INSERT INTO action_table(customer_id,action,request_time) 
VALUES 
('A','LOGIN','2023-05-01'),
('A','SEARCH','2023-05-02'),
('A','SEARCH','2023-05-03'),
('A','ORDER','2023-05-04'),
('B','LOGIN','2023-05-01'),
('B','SEARCH','2023-05-02'),
('B','SEARCH','2023-05-03'),
('B','LOGIN','2023-05-04'),
('B','SEARCH','2023-05-05')

<表类=“s-表”> <标题> customer_id Action 请求时间 <正文> 一个 登录 2023-05-01 一个 搜索 2023-05-02 一个 搜索 2023-05-03 一个 订单 2023-05-04 B 登录 2023-05-01 B 搜索 2023-05-02 B 搜索 2023-05-03 B 登录 2023-05-04 B 搜索 2023-05-05

在本例中，NO 放弃了对 A 的搜索，3 放弃了对 B 的搜索。

这是我的代码:

select customer_id, count(1)
from action_table c1
left join 
(
  select customer_id, action, request_time 
  from action_table
  where action = 'LOGIN'
) c2
  on c1.customer_id = c2.customer_id
  and c2.request_time > c1.request_time
left join 
(
  select customer_id, action, request_time 
  from action_table
  where action = 'ORDER'
) c3
  on c1.customer_id = c3.customer_id
  and c3.request_time > c1.request_time
  and c3.request_time < c2.request_time
where c1.action = 'SEARCH'
  and c2.customer_id IS NOT NULL
  and c3.customer_id IS NULL
group by 1

这是多余的，而且似乎效果不佳:(

Answer 1

一个选项使用LEAD；这个想法是检查每个登录事件的以下两个操作以识别放弃的搜索:

select customer_id, count(*) cnt_abandonned_search
from (
    select a.*, 
        lead(action, 1) over(partition by customer_id order by request_time) lead_action_1,
        lead(action, 2) over(partition by customer_id order by request_time) lead_action_2
    from action_table a
) a
where action = 'LOGIN'                         -- a LOGIN action...
  and lead_action_1 = 'SEARCH'                 -- ... followed by a SEARCH
  and lead_action_2 is distinct from 'ORDER'   -- ... but not followed by an ORDER
group by customer_id

---

如果您想专注于搜索，并获取每个放弃搜索的完整记录，我们可以使用相同的逻辑，但使用 lead 和 lag:

select *
from (
    select a.*, 
        lag (action) over(partition by customer_id order by request_time) lag_action,
        lead(action) over(partition by customer_id order by request_time) lead_action
    from action_table a
) a
where action = 'SEARCH'                    -- a SEARCH action...
  and lag_action = 'LOGIN'                 -- ... preceded by a LOGIN
  and lead_action is distinct from 'ORDER' -- ... but not followed by an ORDER

<强> Demo on DB Fiddle