Hi,
I am using the standard JMP, not the Pro one...
And I would like to build a next best product category model. For example if a customer has bought Product category A, what's the next product they are more likely to purchase Is it feasable in the JMP version I am using? I only have 6 product categories, so nothink like Market Basket Analysis
Your help would be much more appreciated as it is quite urgent.
Thank you so much
Thank You
What does your data look like? Do you have a "basket" for each customer?
I think a possible simple solution could be to start with each basket as a row of data. Then you would have indicator columns for each product category: 1 = puchased; 0 = not purchased.
Then the multivariate correlations (Analyze > Multivariate Methods > Multivariate) of these indicator variables will tell you about which product categories were purchased together.
Not sure what the best way is, maybe the contingency platform, but here's an idea:
I'm guessing you don't have enough data on each customer to make a personalized prediction, so this idea uses the data for all the customers to make a guess. I built a table of synthesized data with a sequence of purchases, like this:
Customer purchase data
And make suggestions like this:
last purchase: 1 suggests next purchase: 2
last purchase: 2 suggests next purchase: 3
last purchase: 3 suggests next purchase: 4
last purchase: 4 suggests next purchase: 5
last purchase: 5 suggests next purchase: 6
last purchase: 6 suggests next purchase: 1
totalProds = 6;
maxPurchases = 10;
totalCustomers = 1000;
_p = 1; // customers buy the next product in the sequence 10% of the time and a random product 90% ...
nextprod = Function( {}, // this is a synthetic model, just to have some data to play with
If( Random Uniform() > .90, // this is a weak relationship
_p = _p + 1;
If( _p > totalProds, _p = 1 );//
, // else
_p = Random Integer( 1, totalProds )
);
Return( _p );
);
// customers have up to maxPurchases, represented as a list of product numbers (starting at 1)
customerHistory = Function( {},
{result = {}, i},
For( i = Random Uniform( 1, maxPurchases ), i > 0, i -= 1,
Insert Into( result, nextprod() )
);
Return( result ); // {4, 5, 5, 1, 2, 4, 4, 6}
);
// make a customer table with synthetic data
dt = New Table( "Untitled",
Add Rows( totalCustomers ),
New Column( "customer", Character, "Nominal", Formula( Right( Char( 1e9 + Row() ), 8 ) ) ),
New Column( "purchases", Expression, "Nominal", Formula( customerHistory() ) ),
);
dt << runformulas; // make the table formulas ready
dt:customer << deleteformula; // and remove the formulas
dt:purchases << deleteformula; // in case the table needs sorting, etc
// record every pair of purchaces in a matrix
pairs = J( totalProds, totalProds, 0 );
For( irow = 1, irow <= N Rows( dt ), irow += 1,
p = dt:purchases[irow]; // p is a list from the data table
While( N Items( p ) >= 2, // as long as p has at least 2 items
pairs[p[1], p[2]] += 1; // use the first two as indexes to the pairs matrix
Remove From( p, 1 ); // remove the first item from p
);
);
Show( pairs ); // display the matrix
// sumarize the results
For( i = 1, i <= totalProds, i += 1,
Write( "\!nlast purchase: ", i, " suggests next purchase: ", Loc Max( pairs[i, 0] ) )
);
The last line might need a little explanation: pairs[i,0] means row i, all the columns. LocMax is a function that returns the location of the maximum value. Since the product numbers are assigned starting at 1, that location is also the product number that is purchased most often after product i is purchased.
