我有以下数据细节:
表1:Table1
在few records
附近的尺寸很小。
表2:Table2
有50 millions
行。
要求:我需要将table1
中的任何字符串列与table2
匹配,例如name
列到name
,并获得匹配百分比(注释列可以是任何,可能是地址或任何字符串列,在单个单元格中有多个单词)。
样本数据:
create table table1(id int, name varchar(100), address varchar(200));
insert into table1 values(1,'Mario Speedwagon','H No 10 High Street USA');
insert into table1 values(2,'Petey Cruiser Jack','#1 Church Street UK');
insert into table1 values(3,'Anna B Sthesia','#101 No 1 B Block UAE');
insert into table1 values(4,'Paul A Molive','Main Road 12th Cross H No 2 USA');
insert into table1 values(5,'Bob Frapples','H No 20 High Street USA');
create table table2(name varchar(100), address varchar(200), email varchar(100));
insert into table2 values('Speedwagon Mario ','USA, H No 10 High Street','[email protected]');
insert into table2 values('Cruiser Petey Jack','UK #1 Church Street','[email protected]');
insert into table2 values('Sthesia Anna','UAE #101 No 1 B Block','[email protected]');
insert into table2 values('Molive Paul','USA Main Road 12th Cross H No 2','[email protected]');
insert into table2 values('Frapples Bob ','USA H No 20 High Street','[email protected]');
预期结果:
tbl1_Name tbl2_Name Percentage
--------------------------------------------------------
Mario Speedwagon Speedwagon Mario 100
Petey Cruiser Jack Cruiser Petey Jack 100
Anna B Sthesia Sthesia Anna around 80+
Paul A Molive Molive Paul around 80+
Bob Frapples Frapples Bob 100
注意:上面给出的只是要了解的样本数据,我在few records
中有table1
,在50 millions
中有table2
。
我的尝试:
第1步:正如Shnugo所建议的那样,规范化数据并存储在同一个表中。
对于table1:
ALTER TABLE table1 ADD Name_Normal VARCHAR(1000);
GO
--00:00:00 (5 row(s) affected)
UPDATE table1
SET Name_Normal=CAST('<x>' + REPLACE((SELECT LOWER(name) AS [*] FOR XML PATH('')),' ','</x><x>') + '</x>' AS XML)
.query(N'
for $fragment in distinct-values(/x/text())
order by $fragment
return $fragment
').value('.','nvarchar(1000)');
GO
对于table2:
ALTER TABLE table2 ADD Name_Normal VARCHAR(1000);
GO
--01:59:03 (50000000 row(s) affected)
UPDATE table2
SET Name_Normal=CAST('<x>' + REPLACE((SELECT LOWER(name) AS [*] FOR XML PATH('')),' ','</x><x>') + '</x>' AS XML)
.query(N'
for $fragment in distinct-values(/x/text())
order by $fragment
return $fragment
').value('.','nvarchar(1000)');
GO
第2步:使用Levenshtein distance in Microsoft Sql Server创建百分比计算功能
第3步:查询以获取匹配百分比。
--00:00:33 (23456 row(s) affected)
SELECT t.name AS [tbl1_Name],t1.name AS [tbl2_Name],
dbo.ufn_Levenshtein(t.Name_Normal,t1.Name_Normal) percentage
into #TempTable
FROM table2 t
INNER JOIN table1 t1
ON CHARINDEX(SOUNDEX(t.Name_Normal),SOUNDEX(t1.Name_Normal))>0
--00:00:00 (23456 row(s) affected)
SELECT *
FROM #TempTable
WHERE percentage >= 50
order by percentage desc;
结论:获得预期的结果,但它正在使用2 hours
来规范化table2
,如上述查询中的评论所述。有什么建议在step 1
为table2
更好地优化?
您是否尝试过研究DQS(数据质量服务)?取决于您的SQL版本,它附带安装文件。 https://docs.microsoft.com/en-us/sql/data-quality-services/data-matching?view=sql-server-2017