Question

我正在构建一个查询，以通过事件通过平台跟踪用户的生命周期。表 EVENTS 有 3 列 USER_ID、DATE_TIME 和 EVENT_NAME。下面是表格的快照，

下面是我的查询，

SELECT * FROM EVENTS
MATCH_RECOGNIZE
(   PARTITION BY USER_ID
    ORDER BY DATE_TIME
    MEASURES MIN(IFF(EVENT_NAME = 'registration new', DATE_TIME, NULL)) AS REGISTRATION_NEW_TIMESTAMP,
             MIN(IFF(EVENT_NAME = 'registration pending confirm', DATE_TIME, NULL)) AS REGISTRATION_PENDING_CONFIRM_TIMESTAMP,
             MIN(IFF(EVENT_NAME = 'your business information', DATE_TIME, NULL)) AS YOUR_BUSINESS_INFORMATION_TIMESTAMP,
             MIN(IFF(EVENT_NAME = 'your personal information', DATE_TIME, NULL)) AS YOUR_PERSONAL_INFORMATION_TIMESTAMP,
             MIN(IFF(EVENT_NAME = 'qualified', DATE_TIME, NULL)) AS QUALIFIED_TIMESTAMP
  ONE ROW PER MATCH
  PATTERN(STEP_1 ANYTHING* STEP_5)
  DEFINE
        STEP_1 AS EVENT_NAME = 'registration new',
        STEP_2 AS EVENT_NAME = 'registration pending confirm',
        STEP_3 AS EVENT_NAME = 'your business information',
        STEP_4 AS EVENT_NAME = 'your personal information',
        STEP_5 AS EVENT_NAME = 'qualified'
)

我的预期结果，

我现在得到的，

以下是我的要求/警告，

下一个事件的时间戳应大于或等于前一个事件的时间戳（以先到者为准，以便通过漏斗的事件的时间戳相等或不断增加）。这个逻辑的一个很好的例子可以用当前和预期结果的差异来解释，即 REGISTRATION_PENDING_CONFIRM_TIMESTAMP 和 QUALIFIED_TIMESTAMP 列中的值。
并非所有用户都有这 5 个事件，例如，如果 USER_ID 54321 没有/跳过事件“您的个人信息”，则结果必须包含其余步骤的数据（现在，如果用户没有/跳过漏斗中的任何事件，查询不返回任何数据）。我觉得这是因为当用户流中缺少定义为度量的事件时，模式搜索失败。

表中事件的顺序不一致，所以我根据业务/漏斗逻辑在度量部分按顺序定义了事件

Answer 1

这不是一个完整的答案，但至少我在这里帮助定义了示例数据（比截图更好），并介绍了 CLASSIFIER 的用法：

create or replace temp table events as
select $1 user_id, $2 date_time, $3 event_name
from values(1,'2020-11-26 15:24:00','registration new')
, (1,'2021-04-12 18:00:00','registration new')
, (1,'2020-11-26 15:24:00','registration pending confirm')
, (1,'2021-04-12 18:11:00','registration pending confirm')
, (1,'2021-04-18 15:04:00','your personal information')
, (1,'2021-04-22 13:13:00','your personal information')
, (1,'2021-04-13 10:22:00','qualified')
, (1,'2021-04-22 13:13:00','qualified')
;


SELECT * FROM EVENTS
MATCH_RECOGNIZE
(   PARTITION BY USER_ID
    ORDER BY DATE_TIME
 
    MEASURES  classifier as class, MIN(IFF(CLASSIFIER = 'STEP_1', DATE_TIME, NULL)) AS REGISTRATION_NEW_TIMESTAMP,
             MIN(IFF(CLASSIFIER = 'STEP_2', DATE_TIME, NULL)) AS REGISTRATION_PENDING_CONFIRM_TIMESTAMP,
             MIN(IFF(CLASSIFIER = 'STEP_3', DATE_TIME, NULL)) AS YOUR_BUSINESS_INFORMATION_TIMESTAMP,
             MIN(IFF(CLASSIFIER = 'STEP_4', DATE_TIME, NULL)) AS YOUR_PERSONAL_INFORMATION_TIMESTAMP,
             MIN(IFF(CLASSIFIER = 'STEP_5', DATE_TIME, NULL)) AS QUALIFIED_TIMESTAMP
 
  ONE ROW PER MATCH
 -- all rows per match
  PATTERN((step_1 | step_2 | step_3 | step_4 | step_5 | coincidence)*)--(STEP_2 | XX)* (STEP_3 | XXX)* (STEP_4 | XX)* (STEP_5 | XX)*)
  DEFINE
        STEP_1 AS EVENT_NAME = 'registration new',
        STEP_2 AS LAG(DATE_TIME) < DATE_TIME AND EVENT_NAME = 'registration pending confirm' ,
        STEP_3 AS LAG(DATE_TIME) < DATE_TIME AND EVENT_NAME = 'your business information',
        STEP_4 AS LAG(DATE_TIME) < DATE_TIME AND EVENT_NAME = 'your personal information',
        STEP_5 AS EVENT_NAME = 'qualified'
        , COINCIDENCE AS LAG(DATE_TIME) = DATE_TIME
);

在 Snowflake 中使用 SQL 进行漏斗分析

1 个答案: