Get most unique combinations of 2 pictures - mysql

Question

I have a mySQL table that holds n number of pictures.
+------------+--------------+
| picture_id | picture_name |
+------------+--------------+
| 1 | ben.jpg |
| 2 | nick.jpg |
| 3 | mark.jpg |
| 4 | james.jpg |
| .. | ... |
| n | abraham.jpg |
+------------+--------------+
For a web application, i need to display 2 pictures simultaneously where the user can vote for one picture or the other. After voting, the user gets a new set of two pictures.
(application use interface)
+---------------------+--------------------+
| Vorte for picture 1 | Vote for picture 2 |
+---------------------+--------------------+
I would like to avoid displaying the same combinations as much as possible. I can create a helper table that will hold all possible combinations.
+----------------+--------------+--------------+
| combination_id | picture_id_1 | Picture_id_2 |
+----------------+--------------+--------------+
| 1 | 1 | 2 |
| 2 | 1 | 3 |
| 3 | 1 | 4 |
| 4 | 1 | 5 |
| .. | .. | .. |
| (n^2-n)/2 | .. | .. |
+----------------+--------------+--------------+
but for 100 pictures, that would be (100^2 - 100)/2 = 4950 (edit) rows, and with every added picture the table would grow exponentially. (which is not a big issue in todays computing i suppose)
But how do i query this table in a way that the user always sees as little duplicates as possible.
Expected result:
run 1: picture_id's = 4,5 (any numbers between 1 and n)
run 2: picture_id's = 2,7
run 3: picture_id's = 5 and 20
...

Answer 1

DEMO:http://rextester.com/VNWIOA4679 (added 100 pic samples) 2 sec query for 1 user w/o any indexes.
I see no need for a helper table as the data can easily be constructed on the fly with proper indexes. at 1000 pictures you're looking at 499,500 combinations a user could vote upon. still easily managed within a database construct as we operate on a set level, not a record level.
Here's one way assuming my own table structures. I can't think of a more efficient way to store/process the data.
Using this approach as new pictures are added the query will generate a larger and larger combination set but always exclude those on which a user has already voted. no code changes for new pics, no regenerating sets just processing each time the ones a user hasn't made a selection upon.
Create table SO46205797_Pics(
PICID int);
Insert into SO46205797_Pics values (1);
Insert into SO46205797_Pics values (2);
Insert into SO46205797_Pics values (3);
Insert into SO46205797_Pics values (4);
Insert into SO46205797_Pics values (5);
Insert into SO46205797_Pics values (6);
Insert into SO46205797_Pics values (7);
Create table SO46205797_UserPicResults (
USERID int,
PICID int,
PICID2 int,
PICChoiceID int);
Insert into SO46205797_UserPicResults values (1,1,2,1);
Insert into SO46205797_UserPicResults values (1,1,3,1);
Insert into SO46205797_UserPicResults values (1,1,4,4);
magic happens here the above was just data setup.
SELECT A.PICID, B.PICID, C.PICChoiceID
FROM SO46205797_Pics A
INNER JOIN SO46205797_Pics B
on A.PICID < B.PICID
LEFT JOIN SO46205797_UserPicResults C
on A.PICID = C.PicID
and B.PICID = C.PICID2
and C.USERID = 1
WHERE C.userID is null;
Note that if we eliminate the C.userID is null part then we see all of the possible combinations (for user1) (note I treat ID 1, 2 the same as ID 2,1 which I think youw ant) for the 2 photos and which ones the user has selected. Since we don't want to display that choice again, we use the c.userID is null to exclude combinations the user already made a choice for.
Also when saving data to the userPicResults, you need to ensure PICID1 is always less than PICID2.
A different way to do this is using a not exists which may be slightly faster.
obviously indexes on USERID, PICID, PICID2 and in that order would be beneficial (i'd probably make it the a combined PK) for SO46205797_UserPicResults and an index on PICID for SO46205797_Pics as the PK.
SELECT A.PICID, B.PICID
FROM SO46205797_Pics A
INNER JOIN SO46205797_Pics B
on A.PICID < B.PICID
WHERE not exists (SELECT *
FROM SO46205797_UserPicResults C
WHERE A.PICID = C.PicID
and B.PICID = C.PICID2
and C.USERID = 1);
I considered maintaining a parent/child relationship for each image for each user; but this approach doesn't store the choices for all combinations.

Answer 2

The goal of this application is to let people vote for one picture against another, right? Then you need to have some kind of vote results table:
vote_results:
| vote_id | user_id | vote_up_picture_id | vote_down_picture_id | ...
Then, based on data from this table you can easily show to a user picture pairs, which he haven't seen yet:
select first.picture_id, second.picture_id
from pictures as first, pictures as second
where not exists(
select * from vote_results v
where (v.vote_up_picture_id = first.picture_id and v.vote_down_picture_id = second.picture_id)
or (v.vote_up_picture_id = second.picture_id and v.vote_down_picture_id = first.picture_id)
) and first.picture_id != second.picture_id
order by rand()
limit 1
PS. As you see, there is no need in helper table with combination_id