Options
FPGrowth
Contributor I
Hello,
I am starting to work with RapidMiner and FPGrowth. I've seen in other messages of the forum that the input format for FPGrowth, is in this way:
-Each row represent a transaction
-Each column represent a nominal attribute
So, I deduce from it that FPGrowth does an intra-transaction analysis not inter-transaction.
My case is a little different, I have clients, and for each client I have a sequence of transactions along the time (Each transaction consists of a group of items). What I want is to detect associations inter transactions, for example, that a lot of clients that buy milk and bread in a transaction, after two transactions also bougth biscuits and butter.
Is there any way to represent this data and to use FPGrowth for this associations?
Thank You very much
I am starting to work with RapidMiner and FPGrowth. I've seen in other messages of the forum that the input format for FPGrowth, is in this way:
-Each row represent a transaction
-Each column represent a nominal attribute
So, I deduce from it that FPGrowth does an intra-transaction analysis not inter-transaction.
My case is a little different, I have clients, and for each client I have a sequence of transactions along the time (Each transaction consists of a group of items). What I want is to detect associations inter transactions, for example, that a lot of clients that buy milk and bread in a transaction, after two transactions also bougth biscuits and butter.
Is there any way to represent this data and to use FPGrowth for this associations?
Thank You very much
0
Answers
the task you are described is different from what FPGrowth was designed for, as you already mentioned. There is aAn algorithm solving exactly your problem: GeneralizedSequentialPatterns. For example it takes into account the time gap between two transactions, for deciding if they can be viewed as one transaction and so on. We have this one available, but only inside a commercial plugin. We might send you a quote for this, if you are interested, just send me a personal message with your address.
Otherwise you might aggregate each transaction, so that the items are collected from all transactions of one customer. But you then will loose all information about the different transactions. So you don't see if one item is part of the same transaction or in the next.
Greetings,
Sebastian