BigQuery 超算交易

Posted 2023-03-24

【中文标题】BigQuery 超算交易

【英文标题】：BigQuery over-counting transactions

【发布时间】：2018-05-11 21:57:09

【问题描述】：

我正在尝试获取指标（会话、用户、新用户、进入、跳出）以及特定来源、广告类型和自定义维度的交易。返回的会话似乎是正确的，但事务是重复的。我不确定如何将点击级别的自定义维度与会话级别的指标混合使用。

任何想法如何更改查询，以免重复值？

#Standard-SQL
SELECT
  trafficSource.source AS source,
  trafficSource.adContent AS adContent,
  (CASE WHEN hcd.index = 10 THEN hcd.value END) Destination,
  COUNT(DISTINCT CONCAT(fullVisitorId, CAST(visitStartTime AS STRING) )) Sessions,
  COUNT(totals.transactions) transactions
FROM `cc-bigquery-venture.534939.ga_sessions_*`,
UNNEST(hits) as hit,
UNNEST(hit.customdimensions) as hcd
WHERE
_TABLE_SUFFIX BETWEEN '20170101' AND '20170131'
GROUP BY
  source,
  adContent,
  Destination
HAVING 
  Destination IS NOT NULL
AND
  Destination !=''
AND 
  source = 'dfa' AND adContent = 'vacationcore'
ORDER BY
  Destination ASC

【参考方案1】：

BigQuery 标准 SQL 如下

#StandardSQL
SELECT
  trafficSource.source AS source,
  trafficSource.adContent AS adContent,
  (
    SELECT ANY_VALUE(hcd.value) 
    FROM UNNEST(hits) AS hit,
      UNNEST(hit.customdimensions) AS hcd
    WHERE hcd.index = 10
  ) Destination,
  COUNT(DISTINCT CONCAT(fullVisitorId, CAST(visitStartTime AS STRING) )) Sessions,
  SUM(totals.transactions) transactions
FROM `cc-bigquery-venture.534939.ga_sessions_*`
WHERE _TABLE_SUFFIX BETWEEN '20170101' AND '20170131'
GROUP BY source, adContent, Destination
HAVING Destination IS NOT NULL
AND Destination !=''
AND source = 'dfa' AND adContent = 'vacationcore'
ORDER BY Destination ASC

【讨论】：

@ElliottBrossard - 谢谢 Elliott！我认为我最初把它放在那里的原因 - 是在结果中显示 0（零）而不是 NULL - 对于给定组根本没有任何事务的情况。但最好不要在 SUM 内部，而是在外部 :o)

啊，明白了。无论如何，很好的答案:) OP 最大的收获应该是对数组的子选择，这可以在会话级别进行分析。