在SQL SERVER数据库中进行数据挖掘，找到最可能的组合

Question

I have to build a query for analyze the "trending" sales from a store. Basically I need to get the occurrences when a combination of articles are bought, for example: When the article 0001 is bought it is very probably that the article 0002 is also bought, so I would like to retrieve something like:

article a | article b | occurrences |
--------- | --------- | ----------- |
0001      | 0002      | 1
0001      | 0003      | 0

In fact I have a table TicketDetails where it is stored every ticket and the article codes contained on each ticket, something like:

store | station | document | consecutive | article
----- | ------- | -------- | ----------- | ------
w     | x       | y        | a           | 0001
w     | x       | y        | a           | 0002 (same ticket, different article)
w     | x       | y        | b           | 0003

Please give me any suggestions on how to build this query, I feel kinda of lost.

Note: As shown above every ticket is a combination of the first 4 columns wxyz

Answer 1

I think you just want a self-join. If you want all articles -- rather than those that just co-occur in an order -- then the SQL is a tad bit trickier.

Let me assume you have a table called articles , so you can first generate all pairs:

select a1.article, a2.article, count(td2.article) as occurrences
from articles a1 join
     articles a2
     on a1.article < a2.article left join -- (a, b) is the same as (b, a)
     ticketDetails td1
     on td1.article = a1.article left join
     ticketDetails td2
     on td2.article = a2.article and
        td2.store = td1.store and
        td2.station = td1.station and
        td2.document = td1.document and
        td2.consecutive = td1.consecutive
group by a1.article, a2.article;

Answer 2

Join TicketDetails to itself, matching tickets, but different articles

select t1.article
      ,t2.article
      ,Count(t1.article)
from ticketdetails t1
left join ticketdetails t2
   on t1.store = t2.store
      t1.station = t2.station
      t1.document = t2.document
      t1.consecutive = t2.consecutive
      t1.article < t2.article
group by t1.article, t2.article