如何找到第二次登录时间之前的事件总数

Question

In BigQuery, I'm trying to find the total number of events before the second login time.

For different user ids, I have multiple events such as "scroll," "user engagement," "log in," "first_visit," "sign_up" etc. For simplicity, let's consider the above as all the events.

For instance, for user_id 2, I have the following information extracted from the raw data (this is a snapshot of the table).

User_id	Event_name	EventTime
2	scroll	2022-10-31 12:28:35
2	sign_up	2022-10-29 08:11:29
2	login	2022-11-01 16:46:34
2	first_visit	2022-10-30 10:45:22
2	login	2022-11-04 08:10:38
2	scroll	2022-11-05 11:18:35
2	user engagement	2022-11-06 08:45:17
2	user engagement	2022-11-07 05:27:32

First, I found the second login time for each user id.

WITH cte AS (
    SELECT *, RANK() OVER (PARTITION BY User_id ORDER BY LoginTime) rnk
    FROM MyData
)

SELECT User_id, LoginTime AS SecondLoginTime
FROM cte
WHERE rnk = 2
ORDER BY User_id;

User_id	SecondLoginTime
1	2022-11-07 09:52:27
2	2022-11-04 08:10:38

I wanted to write a query where I can compare this login time to each event time (for all user ids)and count the events before the SecondLoginTime.

For instance, for user_id 2, I want to make the following comparison:

User_id	Event_name	EventTime	SecondLoginTime
2	scroll	2022-10-31 12:28:35	2022-11-04 08:10:38
2	sign_up	2022-10-29 08:11:29	2022-11-04 08:10:38
2	login	2022-11-01 16:46:34	2022-11-04 08:10:38
2	first_visit	2022-10-30 10:45:22	2022-11-04 08:10:38
2	login	2022-11-04 08:10:38	2022-11-04 08:10:38
2	scroll	2022-11-05 11:18:35	2022-11-04 08:10:38
2	user engagement	2022-11-06 08:45:17	2022-11-04 08:10:38
2	user engagement	2022-11-07 05:27:32	2022-11-04 08:10:38

And find the followig result:

User_id	TotalEventsBeforeSecondVisit
2	4

I also want to apply this logic to all user ids.

Is there a way to do this? Please kindly ask me for clarification if there is anything misssing or if the problem is unclear. I'd really appreciate your help.

Answer 1

You might consider below.

WITH cte AS (
  SELECT *,
         NTH_VALUE(IF(Event_name = 'login', EventTime, NULL), 2 IGNORE NULLS) OVER w AS SecondLoginTime,
         LAST_VALUE(IF(Event_name = 'login', EventTime, NULL) IGNORE NULLS) OVER w AS LastLoginTime
    FROM MyData
  WINDOW w AS (PARTITION BY User_id ORDER BY EventTime ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING)
)
SELECT User_id,
       COUNTIF(EventTime < SecondLoginTime) AS TotalEventsBeforeSecondLogin,
       COUNTIF(EventTime < LastLoginTime) AS TotalEventsBeforeLastLogin,
  FROM cte
 GROUP BY 1;

Query results

Answer 2

This should work:

with LI2 AS (
    SELECT User_Id, EventTime as LoginTime, ROW_NUMBER() OVER (PARTITION BY User_id ORDER BY EventTime) Seq
    FROM MyTbl
    where Event_Name='login'
)
SELECT 
    LI2.User_id
  , LI2.LoginTime AS SecondLoginTime
  , COUNT(OE.User_ID) as EventsBefore2ndLogin
FROM LI2
     left join
     MyTbl OE -- other events before 2nd login
     on OE.User_Id=LI2.USer_Id
        and OE.EventTime<LI2.LoginTime
WHERE LI2.Seq = 2
GROUP BY 
    LI2.User_id
  , LI2.LoginTime
ORDER BY LI2.User_id

MyTbl is your table.

Updated to accommodate 'Events before Last Login:

with LogInSeq AS (
    SELECT User_Id
        , max(case when LoginSeq=2 then LoginTime else null end) as SecondLoginTime
        , max(case when RevLoginSeq=1 then LoginTime else null end) as LastLoginTime
    from (
            SELECT User_Id, EventTime as LoginTime
                    , ROW_NUMBER() OVER (PARTITION BY User_id ORDER BY EventTime) LogInSeq
                    , ROW_NUMBER() OVER (PARTITION BY User_id ORDER BY EventTime desc) RevLoginSeq
            FROM MyTbl
            where Event_Name='login'
         ) LI1
    where LoginSeq=2 or RevLoginSeq=1
    group by User_Id
)

SELECT 
    LISeq.User_id
  , LISeq.SecondLoginTime
  , LISeq.LastLoginTime
  , COUNT(case when OE.EventTime < LISeq.SecondLoginTime then OE.User_id else null end) as EventsBefore2ndLogin
  , COUNT(case when OE.EventTime < LISeq.LastLoginTime then OE.User_id else null end) as EventsBeforeLastLogin
FROM LogInSeq as LISeq
     left join
     MyTbl OE -- other events before 2nd login
     on OE.User_Id=LISeq.USer_Id
GROUP BY 
    LISeq.User_id
  , LISeq.SecondLoginTime
  , LISeq.LastLoginTime
ORDER BY LISeq.User_id;

Admittedly I wasn't aware of advanced capabilities in BigQuery, such as the ones demonstrated by @JayTiger's answer, so you can simplify this by using COUNTIF and NTH_VALUE functions (I don't have access to them).