Identifying records based on one or more common fields

Question

Hi all
This should be easy - but going round in circles. I want to be able to group together records based on whether at least one field is the same as the corresponding field in a different record.
So:
Account No Tel No PostCode
ABC123 17654231 TS24 4ZX
DEF987 8764573 NE1 1AL
QWE921 17654321 SR9 8BD
ASD579 24697615 SR9 8BD
So in the above example the first, third and fourth record would all form one group because the first links to the third based on Tel No and the third links to the fourth based on PostCode. The second record will be in a group of its own. I want to output the above (into a new table) with an additional field linking relevant records together (a group number).
My first pass was to simply:
select the minimum group number from the output table where either the Tel No or Phone No match the current record. If no match then create a new record for this record with the next incremental group number.
This would give
Account No Tel No PostCode Group
ABC123 17654231 TS24 4ZX 1
DEF987 8764573 NE1 1AL 2
QWE921 17654321 SR9 8BD 1
ASD579 24697615 SR9 8BD 1
On the surface that looks OK. The problem occurs if for instance the PostCode in the second record is 'SR9 8BD'. The output would be identical to the above - but I want them to all share the same group number as now there is some commonality between all four records. I'm sure I need to do something recursively but just can't get my head around it.
Would really appreciate some help.
Thanks
Ian

Answer 1

Hi, Ian,
Welcome to the forum!
Be sure to read the forum FAQ {message:id=9360002} It contains several tips that can help you get good answers quickly
976560 wrote:
Hi all
This should be easy - but going round in circles.Ironically, the solution below involves NOCYCLE.
I want to be able to group together records based on whether at least one field is the same as the corresponding field in a different record.
So:
Account No Tel No PostCode
ABC123 17654231 TS24 4ZX
DEF987 8764573 NE1 1AL
QWE921 17654321 SR9 8BD
ASD579 24697615 SR9 8BDWhenever you have a problem, please post CREATE TABLE and INSERT statments for all tables inolved. Since this is your fist message, I'll do it for you:
DROP TABLE     table_x;
CREATE TABLE  table_x
(   Account_No     VARCHAR2 (10)     PRIMARY KEY
,   Tel_No      VARCHAR2 (10)
,   Post_Code     VARCHAR2 (10)
INSERT INTO table_x (account_no, tel_no,     post_code)
       VALUES         ('ABC123',       '17654321', 'TS24 4ZX');
INSERT INTO table_x (account_no, tel_no,     post_code)
       VALUES         ('DEF987',       '8764573',  'NE1 1AL');
INSERT INTO table_x (account_no, tel_no,     post_code)
       VALUES         ('QWE921',       '17654321', 'SR9 8BD');
INSERT INTO table_x (account_no, tel_no,     post_code)
       VALUES         ('ASD579',       '24697615', 'SR9 8BD');
COMMIT;
So in the above example the first, third and fourth record would all form one group because the first links to the third based on Tel No Be careful! It looks like you made a typo there. I changed the number ending in '231' to '321'.
and the third links to the fourth based on PostCode. The second record will be in a group of its own. I want to output the above (into a new table) with an additional field linking relevant records together (a group number).
My first pass was to simply:
select the minimum group number from the output table where either the Tel No or Phone No match the current record. If no match then create a new record for this record with the next incremental group number.
This would give
Account No Tel No PostCode Group
ABC123 17654231 TS24 4ZX 1
DEF987 8764573 NE1 1AL 2
QWE921 17654321 SR9 8BD 1
ASD579 24697615 SR9 8BD 1
On the surface that looks OK. The problem occurs if for instance the PostCode in the second record is 'SR9 8BD'. The output would be identical to the above - but I want them to all share the same group number as now there is some commonality between all four records. I'm sure I need to do something recursively but just can't get my head around it.You said it exactly: you need to do something recursively.
What features of Oracle SQL are recursive? CONNECT BY (as shown below) and, starting in Oracle 11.2, recursive WITH clauses.
Would really appreciate some help.Here's one way to do that:
CREATE TABLE     table_y
AS
WITH   got_related_account_no      AS
     SELECT  table_x.*
     ,     CONNECT_BY_ROOT account_no     AS related_account_no
     FROM     table_x
     CONNECT BY NOCYCLE  account_no     != PRIOR account_no
          AND   (         tel_no      = PRIOR tel_no
                OR    post_code      = PRIOR post_code
             -- OR ...   -- other columns as needed
SELECT       account_no, tel_no, post_code
,       DENSE_RANK () OVER (ORDER BY  MIN (related_account_no))
                     AS grp
FROM       got_related_account_no
GROUP BY  account_no, tel_no, post_code
;This will work in Oracle 10.1 and up.
At the end of this, table_y contains:
ACCOUNT_NO TEL_NO     POST_CODE         GRP
ABC123     17654321   TS24 4ZX            1
ASD579     24697615   SR9 8BD             1
DEF987     8764573    NE1 1AL             2
QWE921     17654321   SR9 8BD             1