Bigquery - 根据时差删除行

Question

我在 BigQuery 中有一个表。它可以是任何数据库。

我想根据时间条件删除行。发生的情况是，如果用户点击太快，它会创建一个需要删除的重复条目。但在某些情况下，由于 IP 地址或广告商不同，两个有效线索会非常接近并且需要保留。当行之间的间隔在 4 秒内时，我们需要进行重复数据删除。

我还需要确保如果一行被标记为重复，则以下行不使用重复行时间戳来派生 4 秒标志。

ip_address  datetime                    advertiser  order_number    Comment
34.195.131  2020-07-03 22:45:02.585 UTC homepage    5678            KEEP
34.195.131  2020-07-03 22:45:05.593 UTC homepage    5678            REMOVE - WITHIN 4 SECONDS OF B2
34.195.131  2020-07-03 22:45:08.923 UTC homepage    5678            KEEP - SINCE B3 WAS REMOVED, C4 IS NOW MORE THAN 4 SECONDS FROM B2
34.195.131  2020-07-03 22:45:13.788 UTC homepage    5678            KEEP
34.195.131  2020-07-03 22:45:16.523 UTC homepage    5678            REMOVE - WITHIN 4 SECONDS OF B5
34.195.131  2020-07-03 22:45:20.393 UTC homepage    5678            KEEP - SINCE B6 WAS REMOVED, LESS THAN 4 SECONDS OF B4
34.195.131  2020-07-03 22:45:21.247 UTC homepage    5678            REMOVE - WITHIN 4 SECONDS OF B7
34.195.131  2020-07-03 22:45:24.924 UTC homepage    5678            KEEP - SINCE B8 WAS REMOVED AND MORE THAN 4 SECONDS OF B7
34.195.131  2020-07-03 22:45:27.443 UTC homepage    5678            REMOVE - WITHIN 4 SECONDS OF B9
34.195.131  2020-07-03 22:45:30.561 UTC homepage    5678            KEEP - SINCE B10 WAS REMOVED AND MORE THAN 4 SECONDS OF B9
34.195.131  2020-07-03 22:45:32.561 UTC homepage    5678            REMOVE - WITHIN 4 SECONDS OF B11
34.195.131  2020-07-03 22:45:33.935 UTC homepage    5678            REMOVE - WITHIN 4 SECONDS OF B11
34.195.131  2020-07-03 22:45:36.083 UTC homepage    5678            KEEP - SINCE B12 AND B13 WERE REMOVED AND MORE THAN 4 SECONDS OF B11
34.195.132  2020-07-03 22:45:38.849 UTC homepage    5678            KEEP - EVEN THOUGH WITHIN 4 SECONDS OF B14, THIS IS A DIFFERENT IP_ADDRESS
34.195.132  2020-07-03 22:45:38.949 UTC homepage    1234            KEEP - EVEN THOUGH WITHIN 4 SECONDS OF B15 THIS IS A NEW ORDER_NUMBER
 

我曾尝试使用 CTE 和自我加入，但目前没有任何成功。谁能告诉我该怎么做或指示如何进一步进行？

Answer 1

如果您可以添加有关 B2、B3 等的描述，我不太确定要求，我想 cmets 会更清楚地解码所需的逻辑。 无论如何，根据我的理解，我实现了以下逻辑：

创建虚拟表：

WITH 
data as

(
SELECT '34.195.131' ip_address,'2020-07-03 22:45:02.585 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:05.593 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:08.923 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:13.788 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:16.523 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:20.393 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:21.247 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:24.924 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:27.443 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:30.561 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:32.561 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:33.935 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.131' ip_address,'2020-07-03 22:45:36.083 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.132' ip_address,'2020-07-03 22:45:38.849 UTC' datetime,'homepage' advertiser,'5678' order_number
UNION ALL
SELECT '34.195.132' ip_address,'2020-07-03 22:45:38.949 UTC' datetime,'homepage' advertiser,'1234' order_number
),

data_corrected
as
(
SELECT ip_address,CAST(datetime As Timestamp) datetime,advertiser,order_number
From data
)

现在的逻辑是，我使用 Lag 和 Lead 窗口函数来获取后面和前面的值，按 d.ip_address,advertiser,order_number 排序记录，然后计算时间增量。

SELECT d.*, LEAD(datetime)
    OVER (PARTITION BY d.ip_address,advertiser,order_number ORDER BY datetime ASC) AS followed_by_click,
    CASE WHEN TIMESTAMP_DIFF(LEAD(datetime)
    OVER (PARTITION BY d.ip_address,advertiser,order_number ORDER BY datetime ASC),d.datetime  , SECOND)<=4 THEN 'Duplicate' ELSE 'Keep' END delta_followed_by_click,
    LAG(datetime)
    OVER (PARTITION BY d.ip_address,advertiser,order_number ORDER BY datetime ASC) AS preceding_click,
    CASE WHEN TIMESTAMP_DIFF(d.datetime  , LAG(datetime)
    OVER (PARTITION BY d.ip_address,advertiser,order_number ORDER BY datetime ASC), SECOND)<=4 THEN 'Duplicate' ELSE 'Keep' END delta_preceding_click,
    FROM data_corrected d
ORDER BY d.datetime desc

希望这有助于您取得成果。