Question

我正在尝试在Google的BigQuery中完成一项任务，这可能需要逻辑我不确定SQL可以本地处理。

我有两张桌子：

第一个表有一个列，每行是一个小写字
第二个表是一个评论数据库（包括评论，评论本身，时间戳等数据）

我想根据第一个表中单词的出现次数对第二个表中的注释进行排序。

以下是我想要做的基本示例，使用python，使用字母而不是单词......但是你明白了这一点：

words = ['a','b','c','d','e']

comments = ['this is the first sentence', 'this is another comment', 'look another sentence, which is also a comment', 'nope', 'no', 'run']

wordcount = {}

for comment in comments:
    for word in words:
        if word in comment:
            if comment in wordcount:
                wordcount[comment] += 1
            else:
                wordcount[comment] = 1

print(sorted(wordcount.items(), key = lambda k: k[1], reverse=True))

输出：

[('look another sentence, which is also a comment', 3), ('this is another comment', 3), ('this is the first sentence', 2), ('nope', 1)]

到目前为止，我已经看到生成SQL查询的最好的事情是执行以下操作：

SELECT
    COUNT(*)
FROM
    table
WHERE
    comment_col like '%word1%'
    OR comment_col like '%word2%'
    OR ...

但是有超过2000个单词......它感觉不对。有什么提示吗？

Answer 1

以下是BigQuery Standard SQL

#standardSQL
SELECT comment, COUNT(word) AS cnt
FROM comments
JOIN words
ON STRPOS(comment, word) > 0 
GROUP BY comment
-- ORDER BY cnt DESC

如果您愿意，可以使用正则表达式：

#standardSQL
SELECT comment, COUNT(word) AS cnt
FROM comments
JOIN words
ON REGEXP_CONTAINS(comment, word)
GROUP BY comment
-- ORDER BY cnt DESC

您可以使用问题

中的虚拟示例来测试/播放上面的内容

#standardSQL
WITH words AS (
  SELECT word
  FROM UNNEST(['a','b','c','d','e']) word
),
comments AS (
  SELECT comment 
  FROM UNNEST(['this is the first sentence', 'this is another comment', 'look another sentence, which is also a comment', 'nope', 'no', 'run']) comment
)
SELECT comment, COUNT(word) AS cnt
FROM comments
JOIN words
ON STRPOS(comment, word) > 0 
GROUP BY comment
ORDER BY cnt DESC

更新：

有任何快速建议只能进行完整的字符串匹配吗？

#standardSQL
WITH words AS (
  SELECT word
  FROM UNNEST(['a','no','is','d','e']) word
),
comments AS (
  SELECT comment 
  FROM UNNEST(['this is the first sentence', 'this is another comment', 'look another sentence, which is also a comment', 'nope', 'no', 'run']) comment
)
SELECT comment, COUNT(word) AS cnt
FROM comments
JOIN words
ON REGEXP_CONTAINS(comment, CONCAT(r'\b', word, r'\b')) 
GROUP BY comment
ORDER BY cnt DESC

Answer 2

如果我理解得很好，我认为您需要这样的查询：

select comment, count(*) cnt
from comments
join words
  on comment like '% ' + word + ' %'   --this checks for `... word ..`; a word between spaces
  or comment like word + ' %'          --this checks for `word ..`; a word at the start of comment
  or comment like '% ' + word          --this checks for `.. word`; a word at the end of comment
  or comment = word                    --this checks for `word`; whole comment is the word
group by comment
order by count(*) desc

SQL Server Fiddle Demo as a sample

如何计算从一个表到另一个表中的注释的单词的出现次数

2 个答案: