Well this is a way I have been advised to do. Previously, I have tried to use 
apriori but the time and memory cost was too much to process half of items with 
even two transactions.

thus I switched to use a heuristic way such as the one suggested in [2] 

Is there anyway that I can use a small part, say 20x20, of the sparse matrix. 
As ECJ does not have such an option I am trying to write a java code for that.

Am I in the right track?

[2] Ghosh et al. 2010, Mining Frequent Itemsets Using GA