Fix EFM train_set fitting

[hieuddo]

Fix RecSys notebook 7 as discussed here.

There are some output differences, which I think is because of some changes from cornac that exclude positive training items from testing.

[tqtg]

@lthoang please have a look and see if the output is as intended. Thanks

[lthoang]

@hieuddo As we discussed, the output looks fine to me.
In addition, could you please increase the num_top_opinions to 4 or 5 to reflect the differences between the explanations on different aspects of MTER model.

[hieuddo]

I changed num_top_opinions to 4 and kept num_top_aspects at 2 to make it diverse. If we increase they would be identical.