题目来源：快手

一、题目描述

快手平台需优化好友推荐算法，需从用户关注行为中筛选出双向关注的用户对（即互相关注关系），用于分析高互动用户群体的行为特征及构建社交图谱。

样例数据
假设关注关系表 follow 包含以下数据：

from_user	to_user
A	B
B	A
A	C
C	A
B	C

二、实现步骤

步骤1：筛选双向关注用户对

作用：通过自连接，筛选出互相关注的用户对。
SQL逻辑：

sql">select 
    t1.from_user as user_a, 
    t1.to_user as user_b
from 
    follow t1
join 
    follow t2
on 
    t1.from_user = t2.to_user 
    and t1.to_user = t2.from_user;

执行结果：

user_a	user_b
A	B
B	A
A	C
C	A

步骤2：去重处理

作用：确保每对用户只输出一行，避免重复。
SQL逻辑：

sql">select 
    case when t1.from_user < t1.to_user then t1.from_user else t1.to_user end as user_a,
    case when t1.from_user < t1.to_user then t1.to_user else t1.from_user end as user_b
from 
    follow t1
join 
    follow t2
on 
    t1.from_user = t2.to_user 
    and t1.to_user = t2.from_user
where 
    t1.from_user < t1.to_user;

执行结果：

user_a	user_b
A	B
A	C

最终SQL实现

sql">select 
    case when t1.from_user < t1.to_user then t1.from_user else t1.to_user end as user_a,
    case when t1.from_user < t1.to_user then t1.to_user else t1.from_user end as user_b
from 
    follow t1
join 
    follow t2
on 
    t1.from_user = t2.to_user 
    and t1.to_user = t2.from_user
where 
    t1.from_user < t1.to_user;

三、其他方法

sql">select
    least(user_a, user_b) as user_id,
    greatest(user_a, user_b) as friend_id
from follow
group by least(user_a, user_b), greatest(user_a, user_b)
having count(*) >= 2;