Extracting specific strings from text field - sql

Question

I'm working in a SQL Server database. I have a table with a 5 character alphanumeric field which will always be 5 characters. It will always be 5 characters and there will never be special characters. This table has roughly 100K rows.
I have another table with a string field that may or may not contains these characters. This table currently has roughly 2500 possible formats. But those can be both added to and modified. Unfortunately, I don't have access to the data used to determine what should be in the field.
Table1.Model
A1234
B1234
A6485
16849
A4665
99999
Table2.StringField
I have purchased model number A1234 after returning B6485
I have purchased model number 16849 after we thought about 99999
I have purchased model number B1234 before also looking at A1234
I returned A4665 and never purchased anything else
I have no money and don’t buy anything
I am looking to scrape the model numbers from these. I am currently using a case statement which accounts for basically 20 of the possible formats. I add on to the case statement as I find new scenarios that might appear in my data.
pseudo code:
Case when stringfield like 'I have purchased model number%return%'
Then substring(stringfield,30,5) as Model1 and substring(stringfield,52,5) as Model2
When stringfield like 'I have purchased model number%'
Then substring(stringfield,30,5) as Model1 and substring(stringfield,59,5) as Model2
When stringfield like 'I returned%'
Then substring(stringfield,11,5) as Model1 and 'N/A' as Model2
Else 'N/A' as Model1 and 'N/A' as Model2
END
Expected results:
I have purchased model number A1234 after returning B6485
Model1 = A1234 Model2=B6485
I have purchased model number 16849 after we thought about 99999
Model1 = 16849 Model2=99999
I have purchased model number B1234 before also looking at A1234
Model1 = B1234 Model2=A1234
I returned A4665 and never purchased anything else
Model1 = A4665 Model2=N/A
I have no money and don’t buy anything
Model1 = N/A Model2=N/A
I am putting the various scenarios into a reference table so that I can just update that as needed.
Is there a better way to do this? It's not a huge deal to just keep an eye on things and make updates as necessary. But it's just one more item on my list of things that needs to be maintained.
Thanks in advance.
One thing that I forgot to mention is that there is sometimes another substring of the field that is like A14351835410571982 - and I don't want anything from that string.
The things that I've thought about trying are:
Crossjoin from Table1 to itself and then saying
If stringfield like '%value1%value2%' then value1 and value2.
But that is 100k x 100k combinations which seems prohibitively large.
Searching stringfield for anything that's 5 characters long followed by a space or a period or a comma that's either all numbers or a single letter and 4 numbers and then somehow getting the first string and the second string in that order.
A combination of the first two: Identify all 5 character strings in all records then crossjoin them and match with wildcards. This would probably be about 20k values instead of 100k
Continuing down the path that I'm currently on and just do it with brute force
** Note: I am a report analyst, not a developer, so I know enough SQL to be dangerous. I can typically follow along with up to mid-complexity SQL but might need help with anything above that.

Answer 1

Here an example of how a combination of string_split and cross apply can get the models from the strings.
create table Models (
code char(5) primary key,
name varchar(30) not null
);
insert into Models (code, name) values
('A1234','Model A4')
, ('B6485','Model B5')
, ('16849','Model 49')
, ('99999','Model Five9')
, ('B1234','Model B4')
, ('A4665','Model A5')
;
create table Comments (
id int identity(1,1) primary key,
comment varchar(max)
);
insert into Comments (comment) values
('I have purchased model number A1234 after returning B6485')
, ('I have purchased model number 16849 after we thought about 99999')
, ('I have purchased model number B1234 before also looking at A1234')
, ('I returned A4665 and never purchased anything')
, ('I have no money and don’t buy anything')
;
create table #tmpCommentCodes (
id int identity(1,1) primary key,
comment_id int,
model_code varchar(max)
);
insert into #tmpCommentCodes (comment_id, model_code)
select c.id, ca.code
from Comments c
cross apply (
select value as code
from string_split(c.comment, ' ') spl
where value COLLATE Latin1_General_BIN like '[A-Z0-9][A-Z0-9][A-Z0-9][A-Z0-9][A-Z0-9]'
) ca;
select tmp.*, model.name
from #tmpCommentCodes as tmp
left join Models as model
on model.code = tmp.model_code
id | comment_id | model_code | name
-: | ---------: | :--------- | :----------
1 | 1 | A1234 | Model A4
2 | 1 | B6485 | Model B5
3 | 2 | 16849 | Model 49
4 | 2 | 99999 | Model Five9
5 | 3 | B1234 | Model B4
6 | 3 | A1234 | Model A4
7 | 4 | A4665 | Model A5
Then the temporary table can be used to replace the codes.
For example:
WITH RCTE_COMMENTS AS
(
SELECT TOP 1 WITH TIES
c.id AS comment_id
, 0 AS lvl
, tmp.id AS tmpId
, REPLACE(c.comment, CONCAT(' ', tmp.model_code), CONCAT(' ', model.name)) AS comment
FROM Comments AS c
LEFT JOIN #tmpCommentCodes AS tmp
ON tmp.comment_id = c.id
LEFT JOIN Models AS model
ON model.code = tmp.model_code
ORDER BY ROW_NUMBER() OVER (PARTITION BY c.id ORDER BY tmp.id)
UNION ALL
SELECT c.comment_id
, c.lvl+1
, tmp.id
, REPLACE(c.comment, ' '+tmp.model_code, ' '+model.name)
FROM RCTE_COMMENTS AS c
JOIN #tmpCommentCodes AS tmp
ON tmp.comment_id = c.comment_id
AND tmp.id = c.tmpId + 1
JOIN Models AS model
ON model.code = tmp.model_code
)
SELECT TOP 1 WITH TIES comment_id
, REPLACE(comment, 'model number ', '') AS comment
FROM RCTE_COMMENTS
ORDER BY ROW_NUMBER() OVER (PARTITION BY comment_id ORDER BY tmpId DESC)
comment_id | comment
---------: | :-----------------------------------------------------------
1 | I have purchased Model A4 after returning Model B5
2 | I have purchased Model 49 after we thought about Model Five9
3 | I have purchased Model B4 before also looking at Model A4
4 | I returned Model A5 and never purchased anything
5 | I have no money and don’t buy anything
db<>fiddle here