Get users who replied the first message of the conversation within 24hours - mysql

Question

I need to find the users who replied the first message of the conversation (one to one conversation) within 24hours. I have a messages table where all data are stored.
Table: messages
id | sender_id | recipient_id | content | Created_at
1 | 1001 | 256 | Hi | 2017-03-20 22:37:30
2 | 256 | 1001 | Hello | 2017-03-21 20:29:10
3 | 1001 | 256 | XYZ | 2017-03-21 22:02:00
4 | 256 | 1001 | ??? | 2017-03-21 23:01:01
5 | 1002 | 500 | Hi there | 2017-03-22 10:10:10
6 | 1002 | 500 | Can you meet?| 2017-03-22 10:15:32
7 | 500 | 1002 | Yes | 2017-03-22 10:20:30
8 | 1003 | 600 | Hello world | 2017-03-23 01:00:00
9 | 1004 | 700 | Hi | 2017-03-23 08:10:10
10 | 700 | 1004 | hello | 2017-03-26 22:00:00
Expected result:
users
256
500
Example: Conversation between user 1001 and 256.
id | sender_id | recipient_id | content | Created_at
1 | 1001 | 256 | Hi | 2017-03-20 22:37:30
2 | 256 | 1001 | Hello | 2017-03-21 20:29:10
3 | 1001 | 256 | XYZ | 2017-03-21 22:02:00
4 | 256 | 1001 | ??? | 2017-03-21 23:01:01
Here 2 | 256 | 1001 | Hello | 2017-03-21 20:29:10 is the first replied message of the conversation and its replied within 24 hours.

Answer 1

I've tested this out and it works. It's much the same as the other answers though.
select messages.sender_id as users from (
select t.id1, t.id2, t.start, messages.sender_id as initiator,
messages.recipient_id as replier from (
select greatest(sender_id, recipient_id) as id1,
least(sender_id, recipient_id) as id2, min(Created_at) as start
from messages group by id1, id2
) as t left join messages on messages.Created_at = t.start
and ((messages.sender_id = t.id1 and messages.receiver_id = t.id2)
or (messages.sender_id = t.id2 and messages.receiver_id = t.id1))
) as t inner join messages on messages.sender_id = t.replier
and messages.recipient_id = t.initiator
and messages.Created_at < date_add(t.start, interval 1 day)
group by users;
The innermost query finds conversations by grouping messages by the two users involved, and finds the start of that conversation by taking the minimum Created_at.
The middle query finds the initiator and replier by looking up the first message in the conversation.
The outside query finds messages from the replier to the initiator (which are therefore in that conversation) within one day of the start of it, and groups by users so that they each appear only once (even if involved in multiple conversations).

Answer 2

Alright.
First, we need to define what a conversation is: a pair of (sender_id, recipient_id) exchanging messages. Determining the first message in a conversation is a bit tricky. We could do this:
SELECT sender_id, recipient_id, min(created_at) FROM messages
GROUP BY sender_id, recipient_id
However, this will give us the first two messages of each conversation instead. We still don't know who started it and who replied without looking at the date, but the data we get is all we need to answer the question. And it is likely to be fast, since I will assume an index on (sender_id, recipient_id, created_at).
Now, I see two ways to solve this. First one:
SELECT least(sender_id,recipient_id),
greatest(sender_id,recipient_id),
max(created_at) <= DATE_ADD( min(created_at), INTERVAL 1 DAY )
FROM (
SELECT sender_id, recipient_id, min(created_at) FROM messages
GROUP BY sender_id, recipient_id
) foo
GROUP BY least(sender_id,recipient_id),
greatest(sender_id,recipient_id)
HAVING count(*)=2;
least() and greatest() allow to create one id for each conversation from the sender and receiver ids. max() and min() will return the first message and its reply, since we have only 2 rows per conversation. And the having will remove messages without reply.
We could also use a temp table:
CREATE TEMPORARY TABLE foo (
sender_id INT NOT NULL,
recipient_id INT NOT NULL,
createdèat DATETIME NOT NULL
);
INSERT INTO foo
SELECT sender_id, recipient_id, min(created_at) FROM messages
GROUP BY sender_id, recipient_id
ALTER TABLE foo ADD PRIMARY KEY (sender_id,recipient_id);
SELECT ... substract a.created_at and b.created_at to get your 24h limit
FROM foo a
JOIN foo b ON ( a.sender_id=b.recipient_id
AND a.recipient_id=b.sender_id
AND a.created_at < b.created_at)
By joining the temp table to itself, we put together the first message and its reply in a single query, and we can compare their dates.

Answer 3

Taking a swing without testing, as I think the desired result is still unclear.
First, find the "first messages" of a conversation:
select m1.id
,m.sender_id
,m.recipient_id
,m.Created_at
from messages m1
inner join (
select m.sender_id
,m.recipient_id
,Min(m.Created_at) as first_message
from messages m
group by m.sender_id
,m.recipient_id
) m2
on m1.sender_id = m2.sender_id
and m1.m.recipient_id = m2m.recipient_id
and m1.Created_at = m2.first_message
If these are properly "first messages", then find any replies in 24 hours
select distinct m3.sender_id
from messages m3
inner join (
<the above first message select statement>
) fm
on m3.sender_id = fm.recipient_id
and m3.recipient_id = fm.sender_id
and m3.Created_at < DATEADD (HH , 24 , fm.Created_at)
where m3.Created_at > fm.Created_at

Answer 4

this return the last message in 24 hours between to users
select
cnv.id ,
cnv.sender_id,
cnv.recipient_id,
cnv.content,
cnv.Created_at
from
(
-- first create a table with costum id of conversaton
select
-- ex: 1001-256
concat(greatest(sender_id, recipient_id),'-',least(sender_id, recipient_id) ) as 'cnv_id', -- costum column for joining
id ,
sender_id,
recipient_id,
content,
Created_at
from message
) cnv
INNER JOIN
(
-- second extract the last date of every one to one coversation conversation
-- result ex : 1001-256 | 2017-03-21 23:01:01
SELECT
concat(greatest(sender_id, recipient_id),'-',least(sender_id, recipient_id) ) as 'cnv_id', -- costum column for joining
max(Created_at) 'max_date'
group by cnv_id
) max_cnv ON cnv.cnv_id = max_cnv.cnv_id -- join the two result by the custom cnv_id
WHERE
-- here we extract only the record that there Created_at is > 24 hours from the max date
-- you can make this condition as you want but i think this will work
(max_cnv.max_date - cnv.Created_at)/1000/60 >= 24;