Trjtdtk

Problem description:

I need to use an association rule algorithm that lets me use database tuples and I think Apriori is a good option, but I am not sure.

Starting point:

I know the Apriori algorithm works in itemsets provided the items are independent one from each other, let's say:

T1: Onions, Beer, Diapers
T2: Chili, Onions, Pizza
T3: Beer, Chili, Onions
...

This would imply building a table of the form:

 -------------------//
 | Onions | Beer
|----------------------//
|T1 | 1 | 1
|T2 | 1 | 0
|----------------------//

Being the rows the transactions, the columns the items and a 0 or a 1 in the cell if the item is in the transaction or not.

Problem details:

I am using a beer database in which I have tuples of different kinds of beer, for example:

 id name type country liters volume container
------------------------------------------------------------------------------
 xxx The Jolly Joker Pale Ale Netherlands Regular Mild Can

...being the upper row the column names and the lower one a sample tuple.

I want to find, given a set of tuples - like a shopping cart of 7 beers -, find rules of association that can fit that cart, as it follows:

Pale Ale, Netherlands -> Bottle

I am doubting if the Apriori algorithm can mine anything on this kind of data structure: one possibility would be treating every single value (regardless the column name) as an item of an itemset and after running the algorithm, ignoring the senseless sets (for example, Netherlands, Germany, Blonde).

I don't know if that approach would compromise the algorithm's reliability, if there exists a better way of getting the rules I need or if I am just using the wrong algorithm and there's one I don't know that fits my specifications.