I am creating a key-wording module where I want to search data using the comma separated words.And the search is categorized into comma , and minus -.
I know a relational database engine is designed from the principle that a cell holds a single value and obeying to this rule can help for performance.But in this case table is already running and have millions of data and can't change the table structure.
Take a look on the example what I exactly want to do is
I have a main table name tbl_main in SQL
AS_ID KWD
1 Man,Businessman,Business,Office,confidence,arms crossed
2 Man,Businessman,Business,Office,laptop,corridor,waiting
3 man,business,mobile phone,mobile,phone
4 Welcome,girl,Greeting,beautiful,bride,celebration,wedding,woman,happiness
5 beautiful,bride,wedding,woman,girl,happiness,mobile phone,talking
6 woman,girl,Digital Tablet,working,sitting,online
7 woman,girl,Digital Tablet,working,smiling,happiness,hand on chin
If search text is = Man,Businessman then result AS_ID is =1,2
If search text is = Man,-Businessman then result AS_ID is =3
If search text is = woman,girl,-Working then result AS_ID is =4,5
If search text is = woman,girl then result AS_ID is =4,5,6,7
What is the best why to do this, Help is much appreciated.Thanks in advance
I think you can easily solve this by creating a FULL TEXT INDEX on your KWD column. Then you can use the CONTAINS query to search for phrases. The FULL TEXT index takes care of the punctuation and ignores the commas automatically.
-- If search text is = Man,Businessman then the query will be
SELECT AS_ID FROM tbl_main
WHERE CONTAINS(KWD, '"Man" AND "Businessman"')
-- If search text is = Man,-Businessman then the query will be
SELECT AS_ID FROM tbl_main
WHERE CONTAINS(KWD, '"Man" AND NOT "Businessman"')
-- If search text is = woman,girl,-Working the query will be
SELECT AS_ID FROM tbl_main
WHERE CONTAINS(KWD, '"woman" AND "girl" AND NOT "working"')
To search the multiple words (like the mobile phone in your case) use the quoted phrases:
SELECT AS_ID FROM tbl_main
WHERE CONTAINS(KWD, '"woman" AND "mobile phone"')
As commented below the quoted phrases are important in all searches to avoid bad searches in the case of e.g. when a search term is "tablet working" and the KWD value is woman,girl,Digital Tablet,working,sitting,online
There is a special case for a single - search term. The NOT cannot be used as the first term in the CONTAINS. Therefore, the query like this should be used:
-- If search text is = -Working the query will be
SELECT AS_ID FROM tbl_main
WHERE NOT CONTAINS(KWD, '"working"')
Here is my attempt using Jeff Moden's DelimitedSplit8k to split the comma-separated values.
First, here is the splitter function (check the article for updates of the script):
CREATE FUNCTION [dbo].[DelimitedSplit8K](
#pString VARCHAR(8000), #pDelimiter CHAR(1)
)
RETURNS TABLE WITH SCHEMABINDING AS
RETURN
WITH E1(N) AS (
SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL
SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1
)
,E2(N) AS (SELECT 1 FROM E1 a, E1 b)
,E4(N) AS (SELECT 1 FROM E2 a, E2 b)
,cteTally(N) AS(
SELECT TOP (ISNULL(DATALENGTH(#pString), 0)) ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) FROM E4
)
,cteStart(N1) AS(
SELECT 1 UNION ALL
SELECT t.N+1 FROM cteTally t WHERE SUBSTRING(#pString, t.N, 1) = #pDelimiter
),
cteLen(N1, L1) AS(
SELECT
s.N1,
ISNULL(NULLIF(CHARINDEX(#pDelimiter, #pString, s.N1),0) - s.N1, 8000)
FROM cteStart s
)
SELECT
ItemNumber = ROW_NUMBER() OVER(ORDER BY l.N1),
Item = SUBSTRING(#pString, l.N1, l.L1)
FROM cteLen l
Here is the complete solution:
-- search parameter
DECLARE #search_text VARCHAR(8000) = 'woman,girl,-working'
-- split comma-separated search parameters
-- items starting in '-' will have a value of 1 for exclude
DECLARE #search_values TABLE(ItemNumber INT, Item VARCHAR(8000), Exclude BIT)
INSERT INTO #search_values
SELECT
ItemNumber,
CASE WHEN LTRIM(RTRIM(Item)) LIKE '-%' THEN LTRIM(RTRIM(STUFF(Item, 1, 1 ,''))) ELSE LTRIM(RTRIM(Item)) END,
CASE WHEN LTRIM(RTRIM(Item)) LIKE '-%' THEN 1 ELSE 0 END
FROM dbo.DelimitedSplit8K(#search_text, ',') s
;WITH CteSplitted AS( -- split each KWD to separate rows
SELECT *
FROM tbl_main t
CROSS APPLY(
SELECT
ItemNumber, Item = LTRIM(RTRIM(Item))
FROM dbo.DelimitedSplit8K(t.KWD, ',')
)x
)
SELECT
cs.AS_ID
FROM CteSplitted cs
INNER JOIN #search_values sv
ON sv.Item = cs.Item
GROUP BY cs.AS_ID
HAVING
-- all parameters should be included (Relational Division with no Remainder)
COUNT(DISTINCT cs.Item) = (SELECT COUNT(DISTINCT Item) FROM #search_values WHERE Exclude = 0)
-- no exclude parameters
AND SUM(CASE WHEN sv.Exclude = 1 THEN 1 ELSE 0 END) = 0
SQL Fiddle
This one uses a solution from the Relational Division with no Remainder problem discussed in this article by Dwain Camps.
From what you've described, you want the keywords that are included in the search text to be a match in the KWD column, and those that are prefixed with a - to be excluded.
Despite the data existing in this format, it still makes most sense to normalize the data, and then query based on the existence or non existence of the keywords.
To do this, in very rough terms:-
Create two additional tables - Keyword and tbl_Main_Keyword. Keyword contains a distinct list of each of the possible keywords and tbl_Main_Keyword contains a link between each record in tbl_Main to each Keyword record where there's a match. Ensure to create an index on the text field for the keyword (e.g. the Keyword.KeywordText column, or whatever you call it), as well as the KeywordID field in the tbl_Main_Keyword table. Create Foreign Keys between tables.
Write some DML (or use a separate program, such as a C# program) to iterate through each record, parsing the text, and inserting each distinct keyword encountered into the Keyword table. Create a relationship to the row for each keyword in the tbl_main record.
Now, for searching, parse out the search text into keywords, and compose a query against the tbl_Main_Keyword table containing both a WHERE KeywordID IN and WHERE KeywordID NOT IN clause, depending on whether there is a match.
Take note to consider whether the case of each keyword is important to your business case, and consider the collation (case sensitive or insensitive) accordingly.
I would prefer cha's solution, but here's another solution:
declare #QueryParts table (q varchar(1000))
insert into #QueryParts values
('woman'),
('girl'),
('-Working')
select AS_ID
from tbl_main
inner join #QueryParts on
(q not like '-%' and ',' + KWD + ',' like '%,' + q + ',%') or
(q like '-%' and ',' + KWD + ',' not like '%,' + substring(q, 2, 1000) + ',%')
group by AS_ID
having COUNT(*) = (select COUNT(*) from #QueryParts)
With such a design, you would have two tables. One that defines the IDs and a subtable that holds the set of keywords per search string.
Likewise, you would transform the search strings into two tables, one for strings that should match and one for negated strings. Assuming that you put this in a stored procedure, these tables would be table-value parameters.
Once you have this set up, the query is simple to write:
SELECT M.AS_ID
FROM tbl_main M
WHERE (SELECT COUNT(*)
FROM tbl_keywords K
WHERE K.AS_ID = M.AS_ID
AND K.KWD IN (SELECT word FROM #searchwords)) =
(SELECT COUNT(*) FROM #searchwords)
AND NOT EXISTS (SELECT *
FROM tbl_keywords K
WHERE K.AS_ID = M.AS_ID
AND K.KWD IN (SELECT word FROM #minuswords))
Related
I have two comma-separated string which needs to be converted into a temptable with two columns synchronized based on the index.
If the input string as below
a = 'abc,def,ghi'
b = 'aaa,bbb,ccc'
then output should be
column1 | column2
------------------
abc | aaa
def | bbb
ghi | ccc
Let us say I have function fnConvertCommaSeparatedStringToColumn which takes in comma-separated string and delimiter as a parameter and returns a column with values. I use this on both strings and get two columns to verify if the count is the same on both sides. But it would be nice two have them in a single temp table. How can i do that?
Let us say I have function which ... returns a column with values.
At that point, the basic idea is to select the column and use the row_number() function with both of your strings. Then you can JOIN the two together using the row_number() result as the matching field for the join.
One method is a recursive CTE:
with cte as (
select convert(varchar(max), null) as a_part, convert(varchar(max), null) as b_part,
convert(varchar(max), 'abc,def,ghi') + ',' as a,
convert(varchar(max), 'aaa,bbb,ccc') + ',' as b,
0 as lev
union all
select convert(varchar(max), left(a, charindex(',', a) - 1)),
convert(varchar(max), left(b, charindex(',', b) - 1)),
stuff(a, 1, charindex(',', a), ''),
stuff(b, 1, charindex(',', b), ''),
lev + 1
from cte
where a <> '' and lev < 10
)
select a_part, b_part
from cte
where lev > 0;
Here is a db<>fiddle.
Here's something a bit sneaky you can try.
I don't have your bespoke function so have used the built-in string_split function (SQL2016+) - for quickly testing, but assuming the parameters are the same. Ideally, your bespoke function should return its own row number in which case you'd use that instead of a rownumber function.
declare #a varchar(20)='abc,def,ghi', #b varchar(20)='aaa,bbb,ccc';
with v as (
select a.value A,b.value B,
row_number() over(partition by a.value order by (select 1/0))Arn,
row_number() over(partition by b.value order by (select 1/0))Brn
from fnConvertCommaSeparatedStringToColumn (#a,',')a
cross apply fnConvertCommaSeparatedStringToColumn (#b,',')b
)
select A,B from v
where Arn=Brn
I would suggest getting a (set based) function that can split a string, based on a delimiter, that returns the ordinal position as well. For example DelimitedSplit8k_LEAD. Then you can trivially split the value, and JOIN on the ordinal position:
DECLARE #a varchar(100) = 'abc,def,ghi';
DECLARE #b varchar(100) = 'aaa,bbb,ccc';
SELECT A.Item AS A,
B.Item AS B
FROM dbo.delimitedsplit8k_lead(#a,',') A
FULL OUTER JOIN dbo.delimitedsplit8k_lead(#a,',') B ON A.ItemNumber = B.ItemNumber;
db<>fiddle
I use a FULL OUTER JOIN and then if either column has a NULL value you know that the 2 delimited lists don't have the same number of delimited values.
I have two columns Reason1, Reason2 both are Non full text search columns and don't have indexes also.
I have to search,
if columns contains keywords like
'Investment','donation','charity','contribution'. and these keywords might contain in Spanish,ITALIAN,... like 7 countries.
How to search Multilanguage keywords against two columns ?
Try the following variant
DECLARE #words varchar(100)='Investment,donation,charity,contribution'
-- put all the words into #words
-- you can use STRING_SPLIT if your version of SQL Server is 2016
;WITH wordsCTE AS(
SELECT
WordList,
1 StartPosition,
CASE WHEN CHARINDEX(',',WordList)>0 THEN CHARINDEX(',',WordList) ELSE LEN(WordList)+1 END CommaPos,
1 WordNum
FROM (SELECT #words WordList) w
UNION ALL
SELECT
i.WordList,
c.CommaPos+1 StartPosition,
CASE WHEN CHARINDEX(',',c.WordList,c.CommaPos+1)>0 THEN CHARINDEX(',',c.WordList,c.CommaPos+1) ELSE LEN(c.WordList)+1 END CommaPos,
c.WordNum+1
FROM (SELECT #words WordList) i
CROSS JOIN wordsCTE c
WHERE c.CommaPos<LEN(c.WordList)
)
SELECT '%'+SUBSTRING(WordList,StartPosition,CommaPos-StartPosition)+'%' word,WordNum
INTO #words
FROM wordsCTE
SELECT *
FROM #words
SELECT *
FROM [your table name is here] d
WHERE EXISTS(
SELECT *
FROM #words w
WHERE (d.Reason1 LIKE w.word OR d.Reason2 LIKE w.word)
)
DROP TABLE #words
But I don't know how it'll be fast in your case.
I have 150,000 rows of data which I'm attempting to query in Google BigQuery.
Column Text contains various lengths of text, from which I want to query for particular keywords.
I've gotten as far as the query below which returns all rows containing a particular keyword (e.g. facebook):
SELECT Text From Data.Set_1
WHERE Text CONTAINS 'facebook'
Questions:
1) How do I improve the query so that it returns a total count of all occurrences of the keyword 'facebook' across 'Text' in a new column?
2) How do I upscale this to multiple keywords (facebook, cnn, bbc, twitter) and return a total count of each keyword present in the data (eg facebook 42, cnn 54, bbc 88, twitter 49)?
for BigQuery Legacy SQL
SELECT
keyword,
COUNT(1) AS rows,
SUM(INTEGER((LENGTH(Text) - LENGTH(REPLACE(Text, keyword, ''))) / LENGTH(keyword))) AS occurences
FROM YourTable
CROSS JOIN keywords
WHERE Text CONTAINS keyword
GROUP BY keyword
Example to play with
SELECT
keyword,
COUNT(1) AS rows,
SUM(INTEGER((LENGTH(Text) - LENGTH(REPLACE(Text, keyword, ''))) / LENGTH(keyword))) AS occurences
FROM (
SELECT Text FROM
(SELECT 'facebookfacebookcnnbbccnn' AS Text),
(SELECT 'facebook' AS Text),
(SELECT 'cnn' AS Text)
) AS words
CROSS JOIN (
SELECT keyword FROM
(SELECT 'facebook' AS keyword),
(SELECT 'cnn' AS keyword),
(SELECT 'bbc' AS keyword)
) AS keywords
WHERE Text CONTAINS keyword
GROUP BY keyword
For BigQuery Standard SQL (see Enabling Standard SQL)
SELECT
keyword,
COUNT(1) AS `rows`,
SUM((LENGTH(Text) - LENGTH(REPLACE(Text, keyword, ''))) / LENGTH(keyword)) AS occurences
FROM YourTable
JOIN keywords
ON STRPOS(Text, keyword) > 0
GROUP BY keyword
Example to play with
WITH keywords AS (
SELECT 'facebook' AS keyword UNION ALL
SELECT 'cnn' AS keyword UNION ALL
SELECT 'bbc' AS keyword
),
words AS (
SELECT 'facebookfacebookcnnbbccnn' AS Text UNION ALL
SELECT 'facebook' AS Text UNION ALL
SELECT 'cnn' AS Text
)
SELECT
keyword,
COUNT(1) AS `rows`,
SUM((LENGTH(Text) - LENGTH(REPLACE(Text, keyword, ''))) / LENGTH(keyword)) AS occurences
FROM words
JOIN keywords
ON STRPOS(Text, keyword) > 0
GROUP BY keyword
You can use a derived table to include all the words you are looking for, and then use aggregation to count the matches:
SELECT w.keyword, COUNT(s.Text)
From (SELECT 'facebook' as keyword UNION ALL
SELECT 'cnn'
) w LEFT JOIN
Data.Set_1 s
ON s.Text CONTAINS w.keyword
GROUP BY w.keyword;
Do note: This is not particularly efficient. The performance should be roughly linear in the number of keywords.
I created this CTE that returns first and last names from 2 different tables. I would like to use the CTE to identify all of the records that have the same last names and the first name of one column starts with the same first letter of another column.
This is an example of the results of the CTE. I want the SELECT using the CTE to return only the highlighted results:
;WITH CTE AS
(
SELECT AD.FirstName AS AD_FirstName, AD.LastName AS AD_LastName, NotInAD.FirstName As NotInAD_FirstName, NotInAD.LastName As NotInAD_LastName
FROM PagingToolActiveDirectoryUsers AD JOIN
(
SELECT FirstName, LastName
FROM #PagingUsersParseName
EXCEPT
SELECT D.FirstName, D.LastName
FROM PagingToolActiveDirectoryUsers D
WHERE D.FirstName <> D.LastName AND D.LastName <> D.LoginName
AND D.LoginName LIKE '%[0-9]%[0-9]%'
) AS NotInAD ON NotInAD.LastName = AD.LastName
)
SELECT *
FROM CTE
WHERE (AD_LastName = NotInAD_LastName) AND (AD_FirstName LIKE ('''' + SUBSTRING(NotInAD_FirstName, 1, 1) + '%'''))
ORDER BY AD_LastName, AD_FirstName;
The result of this query returns no rows.
What am I doing wrong?
Thanks.
You're enclosing the string to be searched for with single-quotes, but it doesn't appear that the data in AD_FirstName has those single-quotes embedded in it. I suggest you replace the first line of the WHERE clause with
WHERE (AD_LastName = NotInAD_LastName) AND (AD_FirstName LIKE (SUBSTRING(NotInAD_FirstName, 1, 1) + '%'))
Best of luck.
I'm looking for a t-sql function to get a string like:
a:b,c:d,e:f
and convert it to a table like
ID Value
a b
c d
e f
Anything I found in Internet incorporated single column parsing (e.g. XMLSplit function variations) but none of them letting me describe my string with two delimiters, one for column separation & the other for row separation.
Can you please guiding me regarding the issue? I have a very limited t-sql knowledge and cannot fork those read-made functions to get two column solution?
You can find a split() function on the web. Then, you can do string logic:
select left(val, charindex(':', val)) as col1,
substring(val, charindex(':', val) + 1, len(val)) as col2
from dbo.split(#str, ';') s(val);
You can use a custom SQL Split function in order to separate data-value columns
Here is a sql split function that you can use on a development system
It returns an ID value that can be helpful to keep id and value together
You need to split twice, first using "," then a second split using ";" character
declare #str nvarchar(100) = 'a:b,c:d,e:f'
select
id = max(id),
value = max(value)
from (
select
rowid,
id = case when id = 1 then val else null end,
value = case when id = 2 then val else null end
from (
select
s.id rowid, t.id, t.val
from (
select * from dbo.Split(#str, ',')
) s
cross apply dbo.Split(s.val, ':') t
) k
) m group by rowid