Question

我有一个包含这些列的表：

UserID1, UserID2, ProductID, PurchaseDate

以下查询在购买表中运行，并返回多个用户之间的交互次数，无论过去31天内的订单如何：

DECLARE @threshold AS INT
DECLARE @days AS INT

SET @threshold = 10
SET @days = 31

SELECT 
    UserID1, UserID2, COUNT(*) AS Counter
FROM 
    (SELECT
        --do this to revert columns and count as one case both Col1,Col2 and Col2,Col1
        CASE 
           WHEN UserID1 < UserID2 
              THEN UserID1 
              ELSE UserID2 
        END AS UserID1,
        CASE 
           WHEN UserID1 < UserID2 
              THEN UserID2 
              ELSE UserID1 
        END AS UserID2
    FROM
        Purchases WITH(NOLOCK)
    WHERE 
        Deadline BETWEEN DATEADD(day, -@days, GETDATE()) AND GETDATE()) t
GROUP BY 
    UserID1, UserID2
HAVING 
    COUNT(*) > @threshold

收益率：

UserID1  UserID2  Counter
1        2        10
3        2        5
4        1        8

但是，我想要的是返回一个包含ProductID和PurchaseDate的表格，如下所示

UserID1  UserID2  ProductID  PurchaseDate
1        2        12345      2017-01-18 00:13:52
1        2        5425       2017-01-12 15:10:02
1        2        64362      2017-01-05 10:10:02
..... for the 10 interactions
3        2        25235      2017-01-18 00:13:52
3        2        436346     2017-01-14 00:13:52
..... for the 5 interactions
4        1        23523      2017-01-14 00:13:52
4        1        135135     2017-01-09 00:13:52
..... for the 8 interactions

有没有办法不将第一个查询的结果放在临时表中，然后再次与Purchases表一起加入以查找所有购买内容？

Answer 1

如果我理解正确，那么简单的窗口COUNT将有助于此。

优化器应该足够聪明，可以在一次扫描表中完成。

DECLARE @threshold AS INT;
DECLARE @days AS INT;

SET @threshold = 10;
SET @days = 31;

WITH
CTE_Purchases
AS
(
    SELECT
        --do this to revert columns and count as one case both Col1,Col2 and Col2,Col1
        CASE 
            WHEN UserID1 < UserID2 
            THEN UserID1 
            ELSE UserID2 
        END AS UserID1
        ,CASE 
            WHEN UserID1 < UserID2 
            THEN UserID2 
            ELSE UserID1 
        END AS UserID2
        ,ProductID
        ,PurchaseDate
    FROM
        Purchases
    WHERE 
        Deadline BETWEEN DATEADD(day, -@days, GETDATE()) AND GETDATE()
)
,CTE_Counts
AS
(
    SELECT
        UserID1
        ,UserID2
        ,ProductID
        ,PurchaseDate
        ,COUNT(*) OVER (PARTITION BY UserID1, UserID2) AS Counter
        -- calc COUNT for groups without explicit GROUP BY
    FROM CTE_Purchases
)
SELECT
    UserID1
    ,UserID2
    ,ProductID
    ,PurchaseDate
    ,Counter
FROM CTE_Counts
WHERE Counter > @threshold
-- this filter is instead of your HAVING
;

Answer 2

免责声明：我没有测试过代码，它是在T-SQL IDE之外编写的。 以下代码基于以下假设： UserID1！= UserID2 。

1）我建议使用MAX / MIN值解决方案来处理[Col1，Col2]，方法与[Col2，Col1]相同。它可能会更好地执行并正确处理NULL。您需要SQL Server 2008（或更高版本）才能工作。

SELECT
    (SELECT MAX(usr) FROM (VALUES (UserID1), (UserID2) as User(usr)) as UserID1,
    (SELECT MIN(usr) FROM (VALUES (UserID1), (UserID2) as User(usr)) as UserID2
FROM
    Purchases

2）现在我们需要实际计算它们之间的相互作用，这应该很容易。为了保持代码清洁，我们可以在前面的语句中使用CTE，我在那里添加截止日期过滤器：

;WITH CTE_UserInteractions AS (
    SELECT
        (SELECT MAX(usr) FROM (VALUES (UserID1), (UserID2) as User(usr)) as FirstUser,
        (SELECT MIN(usr) FROM (VALUES (UserID1), (UserID2) as User(usr)) as SecondUser
    FROM
        Purchases
    WHERE
        Deadline BETWEEN DATEADD(day,-@days,GETDATE()) AND GETDATE()
)

SELECT
    FirstUser,
    SecondUser
FROM
    CTE_UserInteractions
GROUP BY
    FirstUser, SecondUser
HAVING
    COUNT(*) > @Threshold

快速注意：人们可能会发现提前计算左边界限边界会对性能产生积极影响。例如，在运行批处理之前，我们可以这样做：

DECLARE @StartDate DATETIME = DATEADD(DAY,-@days,GETDATE())

然后我们可以在WHERE子句中使用@StartDate。

3）最后，我们可以使用CROSS APPLY来获取结果留下的用户“对”的产品和购买列表。如果性能受到影响，我们可以使用子选择（我的解决方案）或预先填充临时表和步骤＃2的结果。

;WITH CTE_UserInteractions AS (
    SELECT
        (SELECT MAX(usr) FROM (VALUES (UserID1), (UserID2) as User(usr)) as FirstUser,
        (SELECT MIN(usr) FROM (VALUES (UserID1), (UserID2) as User(usr)) as SecondUser
    FROM
        Purchases AS p1
    WHERE
        Deadline BETWEEN DATEADD(day,-@days,GETDATE()) AND GETDATE()
)

SELECT
    groupedUsers.FirstUser as UserID1,
    groupedUsers.SecondUser as UserID2,
    products.ProductID,
    products.PurchaseDate
FROM (
    SELECT
        FirstUser,
        SecondUser
    FROM
        CTE_UserInteractions
    GROUP BY
        FirstUser, SecondUser
    HAVING
        COUNT(*) > @Threshold
) groupedUsers
CROSS APPLY (
    SELECT
        ProductID, PurchaseDate
    FROM
        Purchases AS p1
    WHERE
        p1.UserID1 = FirstUser AND p1.UserID2 = SecondUser
    UNION ALL
    SELECT
        ProductID, PurchaseDate
    FROM
        Purchases AS p2
    WHERE
        p2.UserID2 = FirstUser AND p2.UserID1 = SecondUser
) products

在SQL Server中的GROUP BY之后获取组项

2 个答案: