Build a recommendation from DLRN?

Dear all,

I just tested the Facebook code for recommendation. The DLRN tool has good results (Kaggle ads challenge) in predicting whether an ad will be clicked, given the user info.

But I don’t understand, how would an ad recommendation come out of this ? Does one have to make a generative network to find an ad that the customer is susceptible to click ?

Indeed, the DLRN network is a discriminative network, not a generative one.