SQL Server 匹配词组和排序相关性的最佳方法
Posted
技术标签:
【中文标题】SQL Server 匹配词组和排序相关性的最佳方法【英文标题】:SQL Server best method to match word phrases and order relevence 【发布时间】:2011-09-23 12:41:52 【问题描述】:根据参数中单词的数量(计数)/匹配对 sql varchar 列进行排名的最佳方法是什么,有四个不同的唯一标准。这可能不是一个微不足道的问题,但我面临的挑战是使用我的标准根据“最佳匹配”对行进行排序。
列:描述 varchar(100) 参数:@MyParameter varchar(100)
具有此顺序偏好的输出:
完全匹配(整个字符串匹配)- 总是在前 开头(根据匹配的参数长度递减) 在匹配词数相同的情况下,连续词排名更高的词数排名 单词匹配任何地方(不连续)单词可能不完全匹配,因为允许并且很可能单词的部分匹配,应将出租人值应用于部分单词以进行排名但不是关键的(pot 将匹配每个:pot、potter、potholder、depot、depotting for实例)。以其他单词匹配开头的排名应高于没有后续匹配的排名,但这不是交易杀手/超级重要。
我想要一种方法来对列“以”参数中的值“开始”的位置进行排名。假设我有以下字符串:
'This is my value string as a test template to rank on.'
在第一种情况下,我希望在存在最多字数的列/行中获得排名。
在开始时根据出现次数(最佳匹配)排名第二:
'This is my string as a test template to rank on.' - first
'This is my string as a test template to rank on even though not exact.'-second
'This is my string as a test template to rank' - third
'This is my string as a test template to' - next
'This is my string as a test template' - next etc.
其次:(可能是第一组(开头)之后的第二组/第二组数据 - 这是需要的
我想按@MyParameter 中出现在@MyParameter 中的单词数对行进行排名(排序),其中连续单词的排名高于单独的相同计数。
因此,对于上面的示例字符串,'is my string as shown'
的排名将高于'is not my other string as'
,因为连续字符串(单词一起)的“更好匹配”与相同的单词数。匹配度较高的行(出现的字数)将按降序排列最佳匹配。
如果可能,我想在单个查询中执行此操作。
结果中不应出现任何行。
出于性能考虑,表中的行数不会超过 10,000 行。
表中的值是相当静态的,变化不大,但并非完全如此。
我目前无法更改结构,但稍后会考虑(如单词/短语表)
为了使这稍微复杂一些,单词列表位于两个表中 - 但我可以为此创建一个视图,但是一个表结果(较小的列表)应该先于第二个出现,给定相同匹配的较大数据集结果- 这些表以及表中都会有重复项,我只想要不同的值。选择 DISTINCT 并不容易,因为我想返回一个列(sourceTable),这很可能使行不同,在这种情况下,只能从第一个(较小的)表中选择,但需要所有其他列 DISTINCT(不要考虑“独特”评估中的列。
表格中的伪列:
procedureCode VARCHAR(50),
description VARCHAR(100), -- this is the sort/evaluation column
category VARCHAR(50),
relvu VARCHAR(50),
charge VARCHAR(15),
active bit
sourceTable VARCHAR(50) - just shows which table it comes from of the two
不存在像 ID 列这样的唯一索引
匹配不在要排除的第三个表中 SELECT * FROM (select * from tableone where procedureCode not in (select procedureCode from tablethree))
UNION ALL
(select * from tabletwo where procedureCode not in (select procedureCode from tablethree))
编辑:为了解决这个问题,我创建了一个表值参数,如下所示:
0 Gastric Intubation & Aspiration/Lavage, Treatmen
1 Gastric%Intubation%Aspiration%Lavage%Treatmen
2 Gastric%Intubation%Aspiration%Lavage
3 Gastric%Intubation%Aspiration
4 Gastric%Intubation
5 Gastric
6 Intubation%Aspiration%Lavage%Treatmen
7 Intubation%Aspiration%Lavage
8 Intubation%Aspiration
9 Intubation
10 Aspiration%Lavage%Treatmen
11 Aspiration%Lavage
12 Aspiration
13 Lavage%Treatmen
14 Lavage
15 Treatmen
实际短语在第 0 行的位置
这是我目前的尝试:
CREATE PROCEDURE [GetProcedureByDescription]
(
@IncludeMaster BIT,
@ProcedureSearchPhrases CPTFavorite READONLY
)
AS
DECLARE @myIncludeMaster BIT;
SET @myIncludeMaster = @IncludeMaster;
CREATE TABLE #DistinctMatchingCpts
(
procedureCode VARCHAR(50),
description VARCHAR(100),
category VARCHAR(50),
rvu VARCHAR(50),
charge VARCHAR(15),
active VARCHAR(15),
sourceTable VARCHAR(50),
sequenceSet VARCHAR(2)
)
IF @myIncludeMaster = 0
BEGIN -- Excluding master from search
INSERT INTO #DistinctMatchingCpts (sourceTable, procedureCode, description , category ,charge, active, rvu, sequenceSet
)
SELECT DISTINCT sourceTable, procedureCode, description, category ,charge, active, rvu, sequenceSet
FROM (
SELECT TOP 1
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[COMBO])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
''True'' AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''0CPTMore'' AS sourceTable,
''01'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [CPTMORE] AS CPT
ON CPT.[LEVEL] = PP.[LEVEL]
WHERE
(CPT.[COMBO] IS NULL OR CPT.[COMBO] NOT IN (''Editor'',''MOD'',''CATEGORY'',''Types'',''Bundles''))
AND CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
ORDER BY PP.CODE
UNION ALL
SELECT
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[COMBO])) AS category,
LTRIM(RTRIM([CHARGE])) AS charge,
''True'' AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''0CPTMore'' AS sourceTable,
''02'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [CPTMORE] AS CPT
ON CPT.[LEVEL] LIKE PP.[LEVEL] + ''%''
WHERE
(CPT.[COMBO] IS NULL OR CPT.[COMBO] NOT IN (''Editor'',''MOD'',''CATEGORY'',''Types'',''Bundles''))
AND CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
UNION ALL
SELECT
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[COMBO])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
''True'' AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''0CPTMore'' AS sourceTable,
''03'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [CPTMORE] AS CPT
ON CPT.[LEVEL] LIKE ''%'' + PP.[LEVEL] + ''%''
WHERE
(CPT.[COMBO] IS NULL OR CPT.[COMBO] NOT IN (''Editor'',''MOD'',''CATEGORY'',''Types'',''Bundles''))
AND CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
) AS CPTS
ORDER BY
procedureCode, sourceTable, [description]
END -- Excluded master from search
ELSE
BEGIN -- Including master in search, but present favorites before master for each code
-- Get matching procedures, ordered by code, source (favorites first), and description.
-- There probably will be procedures with duplicated code+description, so we will filter
-- duplicates shortly.
INSERT INTO #DistinctMatchingCpts (sourceTable, procedureCode, description , category ,charge, active, rvu, sequenceSet)
SELECT DISTINCT sourceTable, procedureCode, description, category ,charge, active, rvu, sequenceSet
FROM (
SELECT TOP 1
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[COMBO])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
''True'' AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''0CPTMore'' AS sourceTable,
''00'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [CPTMORE] AS CPT
ON CPT.[LEVEL] = PP.[LEVEL]
WHERE
(CPT.[COMBO] IS NULL OR CPT.[COMBO] NOT IN (''Editor'',''MOD'',''CATEGORY'',''Types'',''Bundles''))
AND CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
ORDER BY PP.CODE
UNION ALL
SELECT TOP 1
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[CATEGORY])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
COALESCE(CASE [ACTIVE] WHEN 1 THEN ''True'' WHEN 0 THEN ''False'' WHEN '''' THEN ''False'' ELSE ''False'' END,''True'') AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''2MasterCPT'' AS sourceTable,
''00'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [MASTERCPT] AS CPT
ON CPT.[LEVEL] = PP.[LEVEL]
WHERE
CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
ORDER BY PP.CODE
UNION ALL
SELECT
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[COMBO])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
''True'' AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''0CPTMore'' AS sourceTable,
''01'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [CPTMORE] AS CPT
ON CPT.[LEVEL] = PP.[LEVEL]
WHERE
(CPT.[COMBO] IS NULL OR CPT.[COMBO] NOT IN (''Editor'',''MOD'',''CATEGORY'',''Types'',''Bundles''))
AND CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
UNION ALL
SELECT
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[CATEGORY])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
COALESCE(CASE [ACTIVE] WHEN 1 THEN ''True'' WHEN 0 THEN ''False'' WHEN '''' THEN ''False'' ELSE ''False'' END,''True'') AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''2MasterCPT'' AS sourceTable,
''01'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [MASTERCPT] AS CPT
ON CPT.[LEVEL] = PP.[LEVEL]
WHERE
CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
UNION ALL
SELECT TOP 1
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[COMBO])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
''True'' AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''0CPTMore'' AS sourceTable,
''02'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [CPTMORE] AS CPT
ON CPT.[LEVEL] LIKE PP.[LEVEL] + ''%''
WHERE
(CPT.[COMBO] IS NULL OR CPT.[COMBO] NOT IN (''Editor'',''MOD'',''CATEGORY'',''Types'',''Bundles''))
AND CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
ORDER BY PP.CODE
UNION ALL
SELECT TOP 1
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[CATEGORY])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
COALESCE(CASE [ACTIVE] WHEN 1 THEN ''True'' WHEN 0 THEN ''False'' WHEN '''' THEN ''False'' ELSE ''False'' END,''True'') AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''2MasterCPT'' AS sourceTable,
''02'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [MASTERCPT] AS CPT
ON CPT.[LEVEL] LIKE PP.[LEVEL] + ''%''
WHERE
CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
ORDER BY PP.CODE
UNION ALL
SELECT
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[COMBO])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
''True'' AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''0CPTMore'' AS sourceTable,
''03'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [CPTMORE] AS CPT
ON CPT.[LEVEL] LIKE PP.[LEVEL] + ''%''
WHERE
(CPT.[COMBO] IS NULL OR CPT.[COMBO] NOT IN (''Editor'',''MOD'',''CATEGORY'',''Types'',''Bundles''))
AND CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
UNION ALL
SELECT
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[CATEGORY])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
COALESCE(CASE [ACTIVE] WHEN 1 THEN ''True'' WHEN 0 THEN ''False'' WHEN '''' THEN ''False'' ELSE ''False'' END,''True'') AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''2MasterCPT'' AS sourceTable,
''03'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [MASTERCPT] AS CPT
ON CPT.[LEVEL] LIKE PP.[LEVEL] + ''%''
WHERE
CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
UNION ALL
SELECT
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[COMBO])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
''True'' AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''0CPTMore'' AS sourceTable,
''04'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [CPTMORE] AS CPT
ON CPT.[LEVEL] LIKE ''%'' + PP.[LEVEL] + ''%''
WHERE
(CPT.[COMBO] IS NULL OR CPT.[COMBO] NOT IN (''Editor'',''MOD'',''CATEGORY'',''Types'',''Bundles''))
AND CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
UNION ALL
SELECT
LTRIM(RTRIM(CPT.[CODE])) AS procedureCode,
LTRIM(RTRIM(CPT.[LEVEL])) AS description,
LTRIM(RTRIM(CPT.[CATEGORY])) AS category,
LTRIM(RTRIM(CPT.[CHARGE])) AS charge,
COALESCE(CASE [ACTIVE] WHEN 1 THEN ''True'' WHEN 0 THEN ''False'' WHEN '''' THEN ''False'' ELSE ''False'' END,''True'') AS active,
LTRIM(RTRIM([RVU])) AS rvu,
''2MasterCPT'' AS sourceTable,
''04'' AS sequenceSet
FROM
@ProcedureSearchPhrases PP
INNER JOIN [MASTERCPT] AS CPT
ON CPT.[LEVEL] LIKE ''%'' + PP.[LEVEL] + ''%''
WHERE
CPT.[CODE] IS NOT NULL
AND CPT.[CODE] NOT IN (''0'', '''')
AND CPT.[CODE] NOT IN (SELECT CPTE.[CODE] FROM CPT AS CPTE WHERE CPTE.[CODE] IS NOT NULL)
) AS CPTS
ORDER BY
sequenceSet, sourceTable, [description]
END
/* Final select - uses artificial ordering from the insertion ORDER BY */
SELECT procedureCode, description, category, rvu, charge, active FROM
(
SELECT TOP 500 *-- procedureCode, description, category, rvu, charge, active
FROM #DistinctMatchingCpts
ORDER BY sequenceSet, sourceTable, description
) AS CPTROWS
DROP TABLE #DistinctMatchingCpts
但是,这不符合单词计数的最佳匹配标准(如样本中的第 1 行值),该标准应匹配从该行中找到的最佳(最多)单词计数。
如果有区别,我可以完全控制表值参数的形式/格式。
如果有用,我会将此结果返回给 c# 程序。
【问题讨论】:
这些是否回答了您的问题? 几个答案,一些想法,但没有一个完全足以获得满足标准列表的完整结果集。目前,我正在设计一个似乎正在做我想做的事情的算法原型——一旦我对它进行了全面审查,我将确定它是否是满足这些目标的可行解决方案。 【参考方案1】:您需要能够拆分字符串才能解决此问题。 I prefer the number table approach to split a string in TSQL
为了使我的以下代码(以及我的拆分功能)正常工作,您需要执行此一个时间表设置:
SELECT TOP 10000 IDENTITY(int,1,1) AS Number
INTO Numbers
FROM sys.objects s1
CROSS JOIN sys.objects s2
ALTER TABLE Numbers ADD CONSTRAINT PK_Numbers PRIMARY KEY CLUSTERED (Number)
一旦设置了 Numbers 表,创建这个拆分函数:
CREATE FUNCTION [dbo].[FN_ListToTable]
(
@SplitOn char(1) --REQUIRED, the character to split the @List string on
,@List varchar(8000)--REQUIRED, the list to split apart
)
RETURNS TABLE
AS
RETURN
(
----------------
--SINGLE QUERY-- --this will not return empty rows
----------------
SELECT
ListValue
FROM (SELECT
LTRIM(RTRIM(SUBSTRING(List2, number+1, CHARINDEX(@SplitOn, List2, number+1)-number - 1))) AS ListValue
FROM (
SELECT @SplitOn + @List + @SplitOn AS List2
) AS dt
INNER JOIN Numbers n ON n.Number < LEN(dt.List2)
WHERE SUBSTRING(List2, number, 1) = @SplitOn
) dt2
WHERE ListValue IS NOT NULL AND ListValue!=''
);
GO
您可以随意创建自己的拆分函数,但您仍然需要 Numbers 表才能让我的解决方案发挥作用。
您现在可以轻松地将 CSV 字符串拆分为表格并加入表格:
select * from dbo.FN_ListToTable(',','1,2,3,,,4,5,6777,,,')
输出:
ListValue
-----------------------
1
2
3
4
5
6777
(6 row(s) affected)
现在试试这个:
DECLARE @BaseTable table (RowID int primary key, RowValue varchar(100))
set nocount on
INSERT @BaseTable VALUES ( 1,'The cows came home empty handed')
INSERT @BaseTable VALUES ( 2,'This is my string as a test template to rank') -- third
INSERT @BaseTable VALUES ( 3,'pencil pen paperclip eraser')
INSERT @BaseTable VALUES ( 4,'wow')
INSERT @BaseTable VALUES ( 5,'no dice here')
INSERT @BaseTable VALUES ( 6,'This is my string as a test template to rank on even though not exact.') -- second
INSERT @BaseTable VALUES ( 7,'apple banana pear grape lemon orange kiwi strawberry peach watermellon')
INSERT @BaseTable VALUES ( 8,'This is my string as a test template') -- 5th
INSERT @BaseTable VALUES ( 9,'rat cat bat mat sat fat hat pat ')
INSERT @BaseTable VALUES (10,'house home pool roll')
INSERT @BaseTable VALUES (11,'This is my string as a test template to') -- 4th
INSERT @BaseTable VALUES (12,'talk wisper yell scream sing hum')
INSERT @BaseTable VALUES (13,'This is my string as a test template to rank on.') -- first
INSERT @BaseTable VALUES (14,'aaa bbb ccc ddd eee fff ggg hhh')
INSERT @BaseTable VALUES (15,'three twice three once twice three')
set nocount off
DECLARE @SearchValue varchar(100)
SET @SearchValue='This is my value string as a test template to rank on.'
;WITH SplitBaseTable AS --expand each @BaseTable row into one row per word
(SELECT
b.RowID, b.RowValue, s.ListValue
FROM @BaseTable b
CROSS APPLY dbo.FN_ListToTable(' ',b.RowValue) AS s
)
, WordMatchCount AS --for each @BaseTable row that has has a word in common withe the search string, get the count of matching words
(SELECT
s.RowID,COUNT(*) AS CountOfWordMatch
FROM dbo.FN_ListToTable(' ',@SearchValue) v
INNER JOIN SplitBaseTable s ON v.ListValue=s.ListValue
GROUP BY s.RowID
HAVING COUNT(*)>0
)
, SearchLen AS --get one row for each possible length of the search string
(
SELECT
n.Number,SUBSTRING(@SearchValue,1,n.Number) AS PartialSearchValue
FROM Numbers n
WHERE n.Number<=LEN(@SearchValue)
)
, MatchLen AS --for each @BaseTable row, get the max starting length that matches the search string
(
SELECT
b.RowID,MAX(l.Number) MatchStartLen
FROM @BaseTable b
LEFT OUTER JOIN SearchLen l ON LEFT(b.RowValue,l.Number)=l.PartialSearchValue
GROUP BY b.RowID
)
SELECT --return the final search results
b.RowValue,w.CountOfWordMatch,m.MatchStartLen
FROM @BaseTable b
LEFT OUTER JOIN WordMatchCount w ON b.RowID=w.RowID
LEFT OUTER JOIN MatchLen m ON b.RowID=m.RowID
WHERE w.CountOfWordMatch>0
ORDER BY w.CountOfWordMatch DESC,m.MatchStartLen DESC,LEN(b.RowValue) DESC,b.RowValue ASC
输出:
RowValue CountOfWordMatch MatchStartLen
----------------------------------------------------------------------- ---------------- -------------
This is my string as a test template to rank on. 11 11
This is my string as a test template to rank on even though not exact. 10 11
This is my string as a test template to rank 10 11
This is my string as a test template to 9 11
This is my string as a test template 8 11
(5 row(s) affected)
它对字符串开头的单词匹配略有不同,因为它查看匹配字符串开头的字符数。
一旦你得到这个工作,你可以尝试通过为 SplitBaseTable 创建一些静态索引表来优化它。可能在您的 @BaseTable 上使用触发器。
【讨论】:
这是一个有趣的想法。当前的挑战是:没有一个单词分隔符,如:)(,/=%-][ 都存在于双引号中,并且在双引号中的短语中有些重要:“Level > 9.0%”或“LDL -C 关于“疯狂”的单词/短语拆分规则,您有三个选项:1) 编写一个 CLR 拆分例程来处理所有必要的逻辑。 2) 在您的字符串中插入一个像PRINT CHAR(182)
这样的字符以清楚地识别拆分。 3) 重新设计表格,使每个“短语”已经被分割成自己的行,并且可以根据 ID 和序列号重构它们。至于主键,添加一个标识列,如下所示:blog.sqlauthority.com/2009/05/03/… 并使其成为 PK
很好的解决方案!我正在使用这种方法,并且效果很好。谢谢!但是是否也可以显示具有相似词的结果?我不是在谈论 leventhein 距离。 LIKE 运算符会为我做这件事......但我不知道如何扩展这个解决方案。【参考方案2】:
听起来您正在寻找一种匹配算法,如果不使用存储过程,可能很难创建该算法。从过去的经验来看,有edit distance algorithms(如Levenshtein)在确定相似性方面非常有用。这些返回一个数字,有时是字符串之间的一些差异,您可以在其上创建自己的加权方程来给出分数。然后,您可以为分数创建排名或阈值以降低误报/正数。
【讨论】:
有点像,但是这些术语非常具体,并且只保留了一组有限的单词——所有这些都是具体的,所以“完全匹配”对于我的目的来说就足够了,只需要获得几组精确的“组”根据描述的优先级进行匹配。不过建议很好。 我也可以使用存储过程或任何我需要的东西来获得正确的结果集。 Levenshtein 距离的好建议,但它适用于字母,用于比较术语。现在这是有趣的部分:这是正确的答案:Levenshtein 距离。你想要达到的是 Levenshtein 距离,但不是用字母,而是用文字。我建议如果可能的话,制作一个 CLR 程序集来计算这个距离。 word-Levenshtein 距离为 1 意味着一个词(不是字母)不在它的位置(它不应该在那里,或者它丢失了)。所以你可以很容易地按这个距离订购。【参考方案3】:不久前我有一个类似的问题。我试图回答的问题是在两个不同的列之间匹配了多少单词,并根据匹配单词的最高百分比进行排名。这超出了我的想象,但我从 Martin 那里得到了一个绝妙的答案。
看看他对my question here的回答。
【讨论】:
【参考方案4】:您所有问题的一个答案:使用 sphynx http://sphinxsearch.com 并且不要在 SQL 中解决这个问题。
Sphynx 是开源的,适用于所有数据库和所有操作系统。
这就是 craigslist 所使用的。
这是本文发布时最好的外部全文搜索系统。它将按照您要求的相关性对您的结果进行排序,并且您不需要花哨的 SQL 表或 SQL 过程。试试看。
【讨论】:
有时,无论好坏,您必须使用 SQL 检索记录(例如,如果您还根据其他条件过滤它们,并且如果您将它们链接到相关表)。以上是关于SQL Server 匹配词组和排序相关性的最佳方法的主要内容,如果未能解决你的问题,请参考以下文章
在 SQL Server 中创建和管理全局 Procs 和 UDF 的最佳方法是啥?