mysql - MySQL 中布尔全文搜索中的`~`（波浪号）运算符的行为与 MySQL 开发人员网站中所述不同

Question

我创建了下表fruits-

CREATE TABLE `fruits` (
  `id` tinyint unsigned NOT NULL AUTO_INCREMENT,
  `name` varchar(200) NOT NULL,
  PRIMARY KEY (`id`),
  FULLTEXT KEY `ft_name` (`name`)
) ENGINE=InnoDB AUTO_INCREMENT=7 DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci

然后我在表中输入了以下值fruits-

SELECT * FROM fruits;
+----+---------------+
| id | name          |
+----+---------------+
|  1 | apple, orange |
|  2 | apple, mango  |
|  3 | mango, kiwi   |
|  4 | mango, guava  |
|  5 | apple, banana |
+----+---------------+

现在我运行以下三个 SQL 查询 -

查询一：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+apple' IN BOOLEAN MODE);
+----+---------------+
| id | name          |
+----+---------------+
|  1 | apple, orange |
|  2 | apple, mango  |
|  5 | apple, banana |
+----+---------------+

查询 2：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+apple -orange' IN BOOLEAN MODE);
+----+---------------+
| id | name          |
+----+---------------+
|  2 | apple, mango  |
|  5 | apple, banana |
+----+---------------+

查询 3：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+apple ~orange' IN BOOLEAN MODE);
+----+---------------+
| id | name          |
+----+---------------+
|  1 | apple, orange |
|  2 | apple, mango  |
|  5 | apple, banana |
+----+---------------+

根据 MySQL 开发人员网站，以下是~“布尔全文搜索”中（波浪号）运算符的功能

https://dev.mysql.com/doc/refman/8.0/en/fulltext-boolean.html

'+苹果~Macintosh'

查找包含单词“apple”的行，但如果该行还包含单词“macintosh”，则将其评分低于如果行不包含单词。这比搜索“+apple -macintosh”“更柔和”，因为“macintosh”的存在导致该行根本不返回。

我~在“查询 3”中尝试了（波浪号）运算符，但输出肯定不是预期的。在这里，预期的行为是排id = 1在最后。

PS 我正在使用 MySQL 版本 - 8.0.26-0ubuntu0.20.04.2 for Linux on x86_64 ((Ubuntu))

score 2 · Accepted Answer

尽管在任何地方都没有关于我的答案的文档，但经过彻底的实验，我得出了这个最合乎逻辑的结论 -

'+' 运算符的存在使 '~' 运算符的任何效果无效

我fruits用以下值更新了我的表 -

SELECT * FROM fruits;
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  1 | apple orange watermelon |
|  2 | apple mango pomegranate |
|  3 | apple mango banana      |
|  4 | mango kiwi pomegranate  |
|  5 | mango guava watermelon  |
|  6 | apple banana kiwi       |
+----+-------------------------+

查询一：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('apple mango ~pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  3 | apple mango banana      |
|  1 | apple orange watermelon |
|  5 | mango guava watermelon  |
|  6 | apple banana kiwi       |
|  2 | apple mango pomegranate |
|  4 | mango kiwi pomegranate  |
+----+-------------------------+

查询 2：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('apple ~pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  1 | apple orange watermelon |
|  3 | apple mango banana      |
|  6 | apple banana kiwi       |
|  2 | apple mango pomegranate |
+----+-------------------------+

查询 3：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('mango ~pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  3 | apple mango banana      |
|  5 | mango guava watermelon  |
|  2 | apple mango pomegranate |
|  4 | mango kiwi pomegranate  |
+----+-------------------------+

在这里，在查询 1、2 和 3 中，没有运算符在值之前apple，mango并且~运算符在值之前pomegranate。这确保了包含该单词的行的pomegranate排名低于其他行。

问题 4：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+apple +mango ~pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  2 | apple mango pomegranate |
|  3 | apple mango banana      |
+----+-------------------------+

问题 5：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+apple ~pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  1 | apple orange watermelon |
|  2 | apple mango pomegranate |
|  3 | apple mango banana      |
|  6 | apple banana kiwi       |
+----+-------------------------+

问题 6：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+mango ~pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  2 | apple mango pomegranate |
|  3 | apple mango banana      |
|  4 | mango kiwi pomegranate  |
|  5 | mango guava watermelon  |
+----+-------------------------+

在这里，在查询中， 4、5 和 6+运算符在值之前，apple并且mango和~运算符在值之前pomegranate。显然，运营商的存在+抵消了运营商的任何影响~。

问题 7：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+apple +mango <pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  3 | apple mango banana      |
|  2 | apple mango pomegranate |
+----+-------------------------+

问题 8：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+apple <pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  1 | apple orange watermelon |
|  3 | apple mango banana      |
|  6 | apple banana kiwi       |
|  2 | apple mango pomegranate |
+----+-------------------------+

问题 9：

SELECT id, name FROM fruits
    -> WHERE MATCH(name) AGAINST
    -> ('+mango <pomegranate'
    -> IN BOOLEAN MODE);
+----+-------------------------+
| id | name                    |
+----+-------------------------+
|  3 | apple mango banana      |
|  5 | mango guava watermelon  |
|  2 | apple mango pomegranate |
|  4 | mango kiwi pomegranate  |
+----+-------------------------+

在这里，在查询 7、8 和 9中，+运算符在值之前apple，mango并且<运算符在值之前pomegranate。这确保了包含该单词的行的pomegranate排名低于其他行。

因此，从这里可以推断出 - 如果+存在运算符，则使用<运算符而不是~运算符

更新

经过大量计算，我创建了一个表格，该表格fruits_score_count显示了score每个fruit完成布尔 FULLTEXT 搜索时的结果。

SELECT * FROM fruits_score_count;
+----+-------------+---------------------+----------------------+
| id | fruit_name  | row_numbers_matched | score                |
+----+-------------+---------------------+----------------------+
|  1 | apple       |                   4 | 0.031008131802082062 |
|  2 | banana      |                   2 |  0.22764469683170319 |
|  3 | guava       |                   1 |   0.6055193543434143 |
|  4 | kiwi        |                   2 |  0.22764469683170319 |
|  5 | mango       |                   4 | 0.031008131802082062 |
|  6 | orange      |                   1 |   0.6055193543434143 |
|  7 | pomegranate |                   2 |  0.22764469683170319 |
|  8 | watermelon  |                   2 |  0.22764469683170319 |
+----+-------------+---------------------+----------------------+

查询一：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('apple mango ~pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  3 | apple mango banana      | 0.062016263604164124 |
|  1 | apple orange watermelon | 0.031008131802082062 |
|  5 | mango guava watermelon  | 0.031008131802082062 |
|  6 | apple banana kiwi       | 0.031008131802082062 |
|  2 | apple mango pomegranate |  -0.7103390693664551 |
|  4 | mango kiwi pomegranate  |  -0.7413471937179565 |
+----+-------------------------+----------------------+

查询 2：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('apple ~pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  1 | apple orange watermelon | 0.031008131802082062 |
|  3 | apple mango banana      | 0.031008131802082062 |
|  6 | apple banana kiwi       | 0.031008131802082062 |
|  2 | apple mango pomegranate |  -0.7413471937179565 |
+----+-------------------------+----------------------+

查询 3：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('mango ~pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  3 | apple mango banana      | 0.031008131802082062 |
|  5 | mango guava watermelon  | 0.031008131802082062 |
|  2 | apple mango pomegranate |  -0.7413471937179565 |
|  4 | mango kiwi pomegranate  |  -0.7413471937179565 |
+----+-------------------------+----------------------+

问题 4：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('+apple +mango ~pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  2 | apple mango pomegranate | 0.062016263604164124 |
|  3 | apple mango banana      | 0.062016263604164124 |
+----+-------------------------+----------------------+

问题 5：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('+apple ~pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  1 | apple orange watermelon | 0.031008131802082062 |
|  2 | apple mango pomegranate | 0.031008131802082062 |
|  3 | apple mango banana      | 0.031008131802082062 |
|  6 | apple banana kiwi       | 0.031008131802082062 |
+----+-------------------------+----------------------+

问题 6：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('+mango ~pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  2 | apple mango pomegranate | 0.031008131802082062 |
|  3 | apple mango banana      | 0.031008131802082062 |
|  4 | mango kiwi pomegranate  | 0.031008131802082062 |
|  5 | mango guava watermelon  | 0.031008131802082062 |
+----+-------------------------+----------------------+

问题 7：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('+apple +mango <pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  3 | apple mango banana      | 0.062016263604164124 |
|  2 | apple mango pomegranate |  -0.7103390693664551 |
+----+-------------------------+----------------------+

问题 8：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('+apple <pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  1 | apple orange watermelon | 0.031008131802082062 |
|  3 | apple mango banana      | 0.031008131802082062 |
|  6 | apple banana kiwi       | 0.031008131802082062 |
|  2 | apple mango pomegranate |  -0.7413471937179565 |
+----+-------------------------+----------------------+

问题 9：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('+mango <pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  3 | apple mango banana      | 0.031008131802082062 |
|  5 | mango guava watermelon  | 0.031008131802082062 |
|  2 | apple mango pomegranate |  -0.7413471937179565 |
|  4 | mango kiwi pomegranate  |  -0.7413471937179565 |
+----+-------------------------+----------------------+

此处，查询 1、查询 2、查询 3、查询 7、查询 8、查询 9 的行为符合预期。

但从查询 4、查询 5、查询 6 可以清楚地看出——

在运算符的存在与+运算符的值之前~基本上使值不可见。

仔细观察也发现——

x ~y并且+x <y是等价的

进一步的实验

查询一：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('+mango apple ~pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  3 | apple mango banana      | 0.062016263604164124 |
|  4 | mango kiwi pomegranate  | 0.031008131802082062 |
|  5 | mango guava watermelon  | 0.031008131802082062 |
|  2 | apple mango pomegranate |  -0.7103390693664551 |
+----+-------------------------+----------------------+

第 1 行id = 3获得最高分数，即和的分数之mango和apple。
第 2 行id = 4获得第二个最高分数，即的分数mango。搜索短语+前面是否存在运算符mango无关紧要。~pomegranate
第 3 行与id = 5第 2 行获得相同的分数。但它的位置低于第 2 行，因为当分数相同时，行按的升序排列primary key，这里id是primary key。
第 4 行id = 2得分最低，因此排在最后。这里由于单词apple存在并且在搜索词组中没有+前面的运算符apple，因此~pomegranate在搜索词组中被考虑在内，这大大降低了分数。

查询 2：

SELECT id, name, score FROM
    -> (SELECT id, name, MATCH(name) AGAINST
    -> ('+mango apple <pomegranate' IN BOOLEAN MODE)
    -> AS score FROM fruits ORDER BY score DESC)
    -> AS temp WHERE score != 0;
+----+-------------------------+----------------------+
| id | name                    | score                |
+----+-------------------------+----------------------+
|  3 | apple mango banana      | 0.062016263604164124 |
|  5 | mango guava watermelon  | 0.031008131802082062 |
|  2 | apple mango pomegranate |  -0.7103390693664551 |
|  4 | mango kiwi pomegranate  |  -0.7413471937179565 |
+----+-------------------------+----------------------+

这再次说明了<即使在操作符存在的情况下操作符也生效+。

这进一步强化了我之前的观察，即——

如果+存在运算符，请使用<运算符而不是~运算符

mysql - MySQL 中布尔全文搜索中的`~`（波浪号）运算符的行为与 MySQL 开发人员网站中所述不同

1 回答 1

Related

Reference