I doubt that the calculation of precision and recall is wrong.

The definition of precision is the proportion of the recommendation list (the y_pred) which is contained in the true list(the y_real).

So, I think the calculation of precision should be
precision[i] = (intersection_size / float(len(y_items_pred)))
if len(y_items_pred) else 0.0

and similarly, the calculation of recall should be
recall[i] = (intersection_size / float(len(y_real[i])))
if len(y_real[i]) else 0.0