MySQL 单列 n-gram 拆分和计数

MySQL single column n-gram split and count

给定 MySQL 中的一列字符串(密码)并给定一个值 N,我正在寻找一种 sql 方法来计算每个 n- 的频率gram(长度为 n 的子串)。

把代码放在里面很重要MySQL,因为在我有的其他环境中,它会导致内存溢出。

同时我发现的唯一可行方法是假设字符串的长度有限(合法假设),select分别通过提取不同位置的子字符串,union然后group bycount,像这样(13 个字符中的 9 克):

Select 
    nueve,
    count(*) as density,
    avg(location) as avgloc

From
    (select 
        mid(pass, 1, 9) as nueve, 1 as location
    from
        passdata
    where
        length(pass) >= 9 and length(pass) <= 13 UNION ALL select 
        mid(pass, 2, 9), 2 as location
    from
        passdata
    where
        length(pass) >= 10 and length(pass) <= 13 UNION ALL select 
        mid(pass, 3, 9), 3 as location
    from
        passdata
    where
        length(pass) >= 11 and length(pass) <= 13 UNION ALL select 
        mid(pass, 4, 9), 4 as location
    from
        passdata
    where
        length(pass) >= 12 and length(pass) <= 13 UNION ALL select 
        mid(pass, 5, 9), 5 as location
    from
        passdata
    where
        length(pass) = 13) as nueves
group by nueve
order by density DESC

结果如下所示:

nueve     density avgloc
123456789 1387    2.4564
234567890 193     2.7306
987654321 141     2.0355
password1 111     1.7748
123123123 92      1.913
liverpool 89      1.618
111111111 86      2.2791

其中 nueve 是 9-gram,density 是出现次数,avgloc 是字符串中的平均起始位置

有什么改进查询的建议吗?我也在为其他 n-gram 做同样的事情。

谢谢!

创建一个table,其中包含从 1 到密码最大长度的所有数字。然后你可以加入这个以获得子串位置。

SELECT nueve, COUNT(*) AS density, AVG(location) as avgloc
FROM (
    SELECT MID(p.pass, n.num, @N) AS nueve, n.num AS location
    FROM passdata AS p
    JOIN numbers_table AS n ON LENGTH(p.pass) >= (@N + n.num - 1)
) AS x
GROUP BY nueve
ORDER BY density DESC