Sql Developer - Can you use a case statement in a cursor to return multiple values - sql

Question

I've been working through a task of trying to classify several million rows of data into a variety of different topics. The data involves calls from our customer support, and we're trying to find a way to classify each call into one of 109 topics. Due to the confidentiality of the data I can't disclose any of the actual data, but will try to give a relatable subset of data that other people could compare to.
DATA:
Incident_Number | Call_Description
000123456 | Issue with oranges and apples
000987654 | oranges
004567891 | with apples and kiwis
026589741 | Issue with kiwis
SQL:
select
Incident_Number,
Call_Description,
(case
when call_description like '%oranges%' then oranges
when call_description like '%apples%' then apples
when call_descritpion like '%kiwis%' then 'kiwis'
else 'Unclassified' end) Topic
from DATA
Question
My hope would be to have Incident 000123456 classified as both oranges and apples and Incident 004567891 get classified as apples and kiwis
Desired Output
Incident_Number | Call_Description ......................| Topic
000123456 ........ | Issue with oranges and apples | oranges
000123456 ........ | Issue with oranges and apples | apples
000987654 ........ | oranges ...................................| oranges
004567891 .........| with apples and kiwis............... | apples
004567891 .........| with apples and kiwis............... | kiwis
026589741 .........| Issue with kiwis........................ | kiwis
Wrapup
From my limited knowledge and what I've garnered from research a simple case statement can't do this because it short circuits after finding the first true value. My question is whether or not it is possible to make some alterations to my code OR instead to somehow set up a cursor to run through my initial table and give me the desired output noted above.
I appreciate any help or advice and hope that I've adhered to the rules of this website (which has honestly saved my butt before!)
Regards,
Richard

Answer 1

I use Microsoft SQL Server instead of Oracle, so I'm not sure about the Oracle syntax, but one solution I have used in the past is to create a temporary table:
CREATE GLOBAL TEMPORARY TABLE my_temp_table (
groupName varchar(50)
) ON COMMIT DELETE ROWS;
Insert Into my_temp_table (groupName) VALUES('oranges')
Insert Into my_temp_table (groupName) VALUES('apples')
Insert Into my_temp_table (groupName) VALUES('kiwis')
then I would inner join to the table to duplicate the records:
select
Incident_Number,
Call_Description,
my_temp_table.groupName Topic
from DATA
inner join my_temp_table
on Data.call_description like '%' + my_temp_table.groupName + '%'
One problem with this method is that if a record doesn't fall into any categories, it will be excluded completely.

Answer 2

One option would be to create (physically or virtually) a table of keywords and use that to join to your table of data. Something like
WITH keywords AS (
SELECT 'apples' topic FROM dual UNION
SELECT 'oranges' FROM dual UNION
SELECT 'kiwis' FROM dual
)
SELECT data.incident_number,
data.incident_description,
keywords.topic
FROM data
JOIN keywords
ON( data.incident_description LIKE '%' || keywords.topic || '%' )
This will work but it's not the most efficient or flexible approach in the world. It doesn't handle different forms of the word well (if a description references the singular "apple" for example). It doesn't handle words that appear within other words (if a description talks about "crabapples" for example). And it's doing a relatively slow match based on scanning through the entire incident_description.
An alternative approach would be to use Oracle Text to index your data. That's likely to be a more complex solution but it would be much more flexible and should be more efficient.

Answer 3

My exaple is for MS SQL Server: http://www.sqlfiddle.com/#!3/8e904/2
You could create a table with details and a simple cross join with it
create table data(
Incident_Number varchar(9),
Call_Description varchar(50))
create table detail(detail varchar(20))
insert into data select
'000123456','Issue with oranges and apples'
union select
'000987654', 'oranges '
union select
'004567891', 'with apples and kiwis'
union select
'026589741', 'Issue with kiwis'
insert into detail select 'kiwis'
union select 'oranges'
union select 'apples'
select * from data a
cross join detail b
where a.call_description like '%'||b.detail||'%'
order by a.Incident_Number
Add a new version http://www.sqlfiddle.com/#!3/b3378/1
It manages all incidents identifying those without coincidences with null
select * from data a
left join detail b
on a.call_description like '%'||b.detail||'%'

Answer 4

I have gone for the CURSOR option to loop through the records in the table and update each row with the category information.
Because you mentioned your rules are quite complicated I think each one might need to be hand written in this procedure and not stored in a table.
Sorry this is MSSQL so might need converting a little for Oracle.
/*BEGIN SETUP TEMP DATA*/
CREATE TABLE #tmp1 (
idno int,
title nvarchar(100),
Categories nvarchar(100) --I have added this row to my temp data, you could store this in another table if needed.
)
INSERT INTO #tmp1 (idno, title) VALUES (1, 'problem apples and oranges')
INSERT INTO #tmp1 (idno, title) VALUES (2, 'problem with apples')
INSERT INTO #tmp1 (idno, title) VALUES (3, 'problem with oranges')
INSERT INTO #tmp1 (idno, title) VALUES (4, 'problem with kiwis')
INSERT INTO #tmp1 (idno, title) VALUES (5, 'problem with something')
/*END SETUP TEMP DATA*/
/*SETUP VARIABLES TO USE IN CURSOR*/
DECLARE #idno int,
#title nvarchar(100)
/*DECLARE CURSOR, OPEND IT AND FETCH DATA INTO INTO IT*/
DECLARE incident_cursor CURSOR FOR
SELECT idno, title
FROM #tmp1 --You could add WHERE Categories IS NULL to only update records that have not been processed
OPEN incident_cursor
FETCH NEXT FROM incident_cursor
INTO #idno, #title
/*LOOP THROUGH CURSOR*/
WHILE ##FETCH_STATUS = 0
BEGIN
DECLARE #allCategories nvarchar(100)
SET #allCategories = ''
/*WRITE RULES HERE TO WORK OUT WHETHER CATEGORY NEEDS ADDING*/
IF (#title LIKE '%apples%') BEGIN SET #allCategories = #allCategories + 'Apples ' END
IF (#title LIKE '%oranges%') BEGIN SET #allCategories = #allCategories + 'Oranges ' END
IF (#title LIKE '%kiwis%') BEGIN SET #allCategories = #allCategories + 'Kiwis ' END
IF #allCategories = '' BEGIN SET #allCategories = 'Uncategorised' END
/*UPDATE ORIGINAL TABLE WITH CATEGORY INFORMATION*/
UPDATE #tmp1 SET Categories = #allCategories WHERE idno = #idno
FETCH NEXT FROM incident_cursor
INTO #idno, #title
END
CLOSE incident_cursor;
DEALLOCATE incident_cursor;
/*THIS ARE JUST TO DISPLAY OUTPUT AND CLEAR UP TEST DATA*/
SELECT * FROM #tmp1
DROP TABLE #tmp1
You could set this to run in a scheduled job to update your records as frequently as you see fit.
Might not be a perfect solution for you but hopefully a starting point

Answer 5

Sorry for how long it's taken me to come back to this. I got pulled into some more pressing deliverables. I ended up figuring out how to get it working by storing the queries in one table as records, then running them using an execute immediate command. I've included the code below in case someone can use it in the future.
create table text_kw_search_queries
(
incident_number varchar2(15),
description varchar2(100),
topic_level_1 varchar2(89),
topic_level_2 varchar2(89)
);
DECLARE
V_SQL VARCHAR2(1000);
CURSOR UPTO IS
SELECT TOPIC_level_1, TOPIC_level_2, description FROM tcstopic
;
BEGIN
FOR i IN UPTO LOOP
V_SQL := 'INSERT /*+APPEND PARALLEL(TEXT_KW_SEARCH_LOOP,2)*/ INTO text_kw_search_queries
SELECT
t2.Incident_Number,
t2.description,
t2.TOPIC_level_1,
t2.TOPIC_level_2
FROM (
SELECT
t1.Incident_number,
t1.description,
'''||i.TOPIC_level_1||''' AS TOPIC_level_1,
'''||i.TOPIC_level_2||''' AS TOPIC_level_2,
(CASE WHEN '|| i.description ||' THEN 1 ELSE 0 END) as FLAG
FROM cso_text_query t1
) t2
WHERE t2.FLAG = 1';
EXECUTE IMMEDIATE V_SQL;
DBMS_OUTPUT.PUT_LINE(V_SQL);
COMMIT WORK;
END LOOP;
END;