Hierarchical query to combine two groupings into one broad joint grouping

Question

Hi there,
I would like to know if anyone knows a way to solve the problem below with a SQL querie, maybe using some hierarchical queries or window functions (or anything else in SQL for that matter).
My environment is:
Oracle Database 11g Release 11.2.0.2.0 - 64bit
The problem is this:
I have a list of items that are grouped together in two different grouping ways (two columns).
This gives the ability for items to be linked to other items in two ways:
1. Directly if both have same value on GROUP1 and/or GROUP2;
2. indirectly if they have an item in common with at least one match on either GROUP1 or GROUP2.
The idea is to start from this dataset:
WITH T AS
  SELECT 1 AS ITEM_ID, 'A' AS GROUP1, 100 AS GROUP2 FROM DUAL UNION
  SELECT 2 AS ITEM_ID, 'A' AS GROUP1, 100 AS GROUP2 FROM DUAL UNION
  SELECT 3 AS ITEM_ID, 'A' AS GROUP1, 101 AS GROUP2 FROM DUAL UNION
  SELECT 4 AS ITEM_ID, 'B' AS GROUP1, 100 AS GROUP2 FROM DUAL UNION
  SELECT 5 AS ITEM_ID, 'B' AS GROUP1, 102 AS GROUP2 FROM DUAL UNION
  SELECT 6 AS ITEM_ID, 'C' AS GROUP1, 103 AS GROUP2 FROM DUAL UNION
  SELECT 7 AS ITEM_ID, 'D' AS GROUP1, 101 AS GROUP2 FROM DUAL
SELECT * FROM T;
And end up with this dataset with a one single joint grouping:
WITH T AS
  SELECT 1000 AS JOINT_GROUP_ID, 1 AS ITEM_ID FROM DUAL UNION
  SELECT 1000 AS JOINT_GROUP_ID, 2 AS ITEM_ID FROM DUAL UNION
  SELECT 1000 AS JOINT_GROUP_ID, 3 AS ITEM_ID FROM DUAL UNION
  SELECT 1000 AS JOINT_GROUP_ID, 4 AS ITEM_ID FROM DUAL UNION
  SELECT 1000 AS JOINT_GROUP_ID, 5 AS ITEM_ID FROM DUAL UNION
  SELECT 1000 AS JOINT_GROUP_ID, 7 AS ITEM_ID FROM DUAL UNION
  SELECT 2000 AS JOINT_GROUP_ID, 6 AS ITEM_ID FROM DUAL
SELECT * FROM T;The relationships are:
Item 1 is linked to Item 2 by GROUP1 and GROUP2;
Item 1 is linked to Item 3 by GROUP1 only;
Item 1 is linked to Item 4 by GROUP2 only;
Item 1 is linked to Item 5 through Item 4 by GROUP1;
Item 1 is linked to Item 7 through Item 3 by GROUP2;
Item 6 is not linked to any other item since it does not match on GROUP1 nor GROUP2 with any other item.
NOTEs:
- JOINT_GROUP_ID values could be any sequential value. I used 1000 and 2000 just to avoid confusion with the other IDs and group values used to picture the problem.
- The level of relationship is not restricted to 2 like the example above. There could be deeper relationships.
This seems to me like something that could be solved with a hierarchical query, but I could not get my head around it to solve the problem.
Hope one of you guys can help me on this.
Chears.

Answer 1

Hi Bruno,
You are correct. Frank's solution does not work. You can do this using CONNECT BY on smaller problems, but it will be very inefficient for larger problems wirth significant looping. I wrote a quick blog article on this subject after reading your question this morning. This is actually an example of a very general class of problems and I already had three SQL solutions. I'll put the CONNECT BY one here, and you can look at my blog if you want more details (I include a diagram so I can't just post it here).
Data
item_groups
SQL> SELECT *
  2    FROM item_groups
  3  /
ITEM_ID    GROUP1                         GROUP2
01         A                              100
02         A                              100
03         A                              101
04         B                              100
05         B                              102
06         C                              103
07         D                              101
08         E                              104
09         E                              105
10         F                              106
10 rows selected.Query
WITH links_v AS (
SELECT t_fr.item_id node_id_fr,
       t_to.item_id node_id_to,
       t_fr.item_id || '-' || Row_Number() OVER (PARTITION BY t_fr.item_id ORDER BY t_to.item_id) link_id
  FROM item_groups t_fr
  JOIN item_groups t_to
    ON t_to.item_id > t_fr.item_id
   AND (t_to.group1 = t_fr.group1 OR t_to.group2 = t_fr.group2)
), nodes_v AS (
SELECT item_id node_id
   FROM item_groups
), tree AS (
SELECT link_id, CONNECT_BY_ROOT (link_id) root_id
  FROM links_v
CONNECT BY NOCYCLE (node_id_fr = PRIOR node_id_to OR node_id_to = PRIOR node_id_fr OR
                     node_id_fr = PRIOR node_id_fr OR node_id_to = PRIOR node_id_to)
), group_by_link AS (
SELECT DISTINCT Min (root_id) OVER (PARTITION BY link_id) group_id, link_id
  FROM tree
), linked_nodes AS (
SELECT g.group_id, l.node_id_fr node_id
  FROM group_by_link g
  JOIN links_v l
    ON l.link_id = g.link_id
UNION
SELECT g.group_id, l.node_id_to
  FROM group_by_link g
  JOIN links_v l
    ON l.link_id = g.link_id
SELECT l.group_id "Network", l.node_id "Node"
  FROM linked_nodes l
UNION ALL
SELECT '00 (unlinked)', node_id
  FROM nodes_v n
WHERE n.node_id NOT IN (SELECT node_id FROM linked_nodes)
ORDER BY 1, 2Output
Network       Node
00 (unlinked) 06
              10
01-1          01
              02
              03
              04
              05
              07
08-1          08
              09