Grouping by 2 column - mysql

Question

I have a query that shows how many messages, is being sent through my system the last year, grouped by months. Works perfectly!
The result look like this:
+------+-------+--------+--------+--------+
| Year | Month | Type 1 | Type 2 | Type 3 |
+------+-------+--------+--------+--------+
| 2013 | 10 | 0 | 2 | 3 |
| 2013 | 11 | 4 | 21 | 56 |
| 2013 | 12 | 1 | 10 | 16 |
| 2014 | 1 | 2 | 10 | 52 |
| 2014 | 2 | 1 | 62 | 118 |
+------+-------+--------+--------+--------+
(type 1,2 and 3 is simply different types of USERS -ignore this)
However, I'd like to avoid that the same receiver (msg_receiver) can be shown twice in the result set, for each month.
So if user 44 and 39 sends a message to user 70 in december, user_id 70 would only be counted ONCE for december. Currently, he will show up twice.
Below is my query:
SELECT
Year(m.msg_date) as year,
Month(m.msg_date) as month,
sum(u.type = '1') as type_1,
Sum(u.type = '2') as type_2,
sum(u.type = '7') as type_3
FROM
messages m
INNER JOIN
users u ON u.user_id = m.msg_sender
WHERE
m.msg_date >= CURDATE() - INTERVAL 1 YEAR
AND month(msg_date) != month(curdate())
GROUP BY
Month(m.msg_date) -- , m.msg_receiver (this does not work, it will no longer group by each month/year).
ORDER BY
msg_date
The logical answer to this, would in my option be, to first group by month, then user_id (or vice via). But if I do this, the results looks strange. See:
Using GROUP BY Month(m.msg_date), u.user_id
+------+-------+--------+--------+--------+
| Year | Month | Type 1 | Type 2 | Type 3 |
+------+-------+--------+--------+--------+
| 2013 | 10 | 0 | 1 | 0 |
| 2013 | 10 | 0 | 0 | 1 |
| 2013 | 10 | 0 | 0 | 1 |
| 2013 | 10 | 0 | 1 | 0 |
| 2013 | 10 | 0 | 0 | 1 |
| 2013 | 11 | 0 | 0 | 19 |
| 2013 | 11 | 0 | 1 | 0 |
| 2013 | 11 | 0 | 1 | 0 |
| 2013 | 11 | 0 | 1 | 0 |
| 2013 | 11 | 0 | 1 | 0 |
| 2013 | 11 | 2 | 0 | 0 |
| 2013 | 11 | 0 | 0 | 11 |
+------+-------+--------+--------+--------+
It does not group by months anymore, as it should.
Any ideas?
EDIT
Just to clarify what exactly I want to achieve, as people have been a bit confused. Imagine this scenario:
It is December 2013.
USER 1 has written 5 messages to USER 2 (this should count as 1 in december)
USER 4 has written 1 message to USER 4 (this should count as 1 in december)
USER 3 has written 2 messages to USER 4 and 2 (this should count as 2 in december).
The totals of the month would then be 4. Because there has been 4 conversations.
Does it makes sense? I find my self often struggling with how to express my self correctly and understandable.

Answer 1

You can use COUNT(DISTINCT to only count each msg_receiver once per type:
SELECT
Year(m.msg_date) as year,
Month(m.msg_date) as month,
COUNT(DISTINCT CASE WHEN u.type = '1' THEN m.msg_receiver END) as type_1,
COUNT(DISTINCT CASE WHEN u.type = '2' THEN m.msg_receiver END) as type_2,
COUNT(DISTINCT CASE WHEN u.type = '3' THEN m.msg_receiver END) as type_3
FROM
messages m
INNER JOIN
users u ON u.user_id = m.msg_sender
WHERE
m.msg_date >= CURDATE() - INTERVAL 1 YEAR
AND month(msg_date) != month(curdate())
GROUP BY
Year(m.msg_date), Month(m.msg_date)
ORDER BY
msg_date
N.B I have added Year(m.msg_date) to your group by to ensure the results are determinate
If the same user receives a message from two different users that have two different types, they will be counted in both types though. If this is not the intended result you would need to come up with some logic as to which type they should be counted in (Min, Max, Mode, Median etc)
If, for example, you wanted the minimum user type, you could use:
SELECT
m.year,
m.month,
sum(m.type = '1') as type_1,
Sum(m.type = '2') as type_2,
sum(m.type = '7') as type_3
FROM (
SELECT
Year(m.msg_date) as year,
Month(m.msg_date) as month,
m.msg_receiver,
MIN(u.type) AS type
FROM
messages m
INNER JOIN
users u ON u.user_id = m.msg_sender
WHERE
m.msg_date >= CURDATE() - INTERVAL 1 YEAR
AND month(msg_date) != month(curdate())
GROUP BY
Year(m.msg_date), Month(m.msg_date), m.msg_receiver
) m
GROUP BY
m.Year, m.Month
ORDER BY
m.year, m.month;
EDIT
In response to your updated question, in its current form my first answer would count your example as only 3 conversations not 4, as there were only 3 unique recipients. What you really need is to be able to count distinct over sender and receiver, i.e. count(distinct m.msg_sender, m.msg_sender). Unfortunately this is not valid syntax, however, you can achieve essentially the same thing by concatenating the two fields (as long as they are separated by a character/characters that cannot appear in either. e.g
SELECT
Year(m.msg_date) as year,
Month(m.msg_date) as month,
COUNT(DISTINCT CASE WHEN u.type = '1' THEN CONCAT(m.msg_sender, '|', m.msg_receiver) END) as type_1,
COUNT(DISTINCT CASE WHEN u.type = '2' THEN CONCAT(m.msg_sender, '|', m.msg_receiver) END) as type_2,
COUNT(DISTINCT CASE WHEN u.type = '3' THEN CONCAT(m.msg_sender, '|', m.msg_receiver) END) as type_3
FROM
messages m
INNER JOIN
users u ON u.user_id = m.msg_sender
WHERE
m.msg_date >= CURDATE() - INTERVAL 1 YEAR
AND month(msg_date) != month(curdate())
GROUP BY
Year(m.msg_date), Month(m.msg_date)
ORDER BY
msg_date

Answer 2

you haven't posted a data structure, but it appears that you want to change the INNER JOIN to
INNER JOIN
users u ON u.user_id = m.msg_receiver