如果您不介意某些表数据无法处理(取决于表大小，预计50％的行不能进行处理)，这是一个解决方案:

If you don't mind around some of the table data cannot be processed (depends on table size, expect 50% row cannot be processed), here's a solution:

id  name
1   Some
2   Body
3   Once
4   Told
5   Me
6   The
7   World
8   Is
9   Gonna
10  Roll
11  Me
12  I 
13  Ain't
14  The 
15  Shapest
16  Tool
17  In
18  The
19  Shed
20  She
21  was
22  looking
23  kind
24  of
25  dumb
26  with
27  her
28  finger
29  and
30  her
31  thumb

查询:

SELECT new_id, name FROM (
    SELECT  new_id, name FROM (
        SELECT new_meme.id as new_id, original_meme.id as original_id, original_meme.name FROM meme original_meme
        JOIN meme new_meme ON new_meme.id <>  original_meme.id
        ORDER BY RAND()
    ) layer1
    GROUP BY layer1.new_id
) layer2 GROUP BY name

结果(当然，每次运行都不同)

1   I 
2   In
3   Gonna
4   Ain't
5   The
6   her
7   finger
8   Some
9   dumb
10  She
15  Me
16  with
17  Told
18  and
19  World
21  Roll
22  The 
25  Tool
26  Shed
27  Is
28  Me
29  Sharpest
31  The

注意:您可能会发现查询非常慢，这是因为它两次连接了表，因此，如果数据大小为1000，则需要处理1000 * 1000.

Note: You might find the query very slow, this is because it join the table twice, so if data size is 1000, it will need to process 1000 * 1000.

您可以通过将ON new_meme.id <> original_meme.id切换为ON new_meme.id BETWEEN original_meme.id - 5 AND original_meme.id +5(可以更改5)来控制查询的速度，但是它将降低随机性，并且不适用于非数字ID

You can control the speed of the query by chaging ON new_meme.id <> original_meme.id to ON new_meme.id BETWEEN original_meme.id - 5 AND original_meme.id +5 (the 5 can be change), but it will decrease randomness and not worked for non-numberical id