25.5.1.1 Impact on Recommender and Attack Performance
The misclassification of authentic profiles results in the removal of good data from
the ratings database, which has the potential to impact negatively on the overall performance of the recommender system. One way to assess this impact is to compute
the MAE of the system before and after detection and filtering. On the positive side,
the removal of attack profiles reduces attack performance. Assuming the attack is
a push or nuke attack, the degree to which attack performance is affected can be
assessed by computing the prediction shift on the targeted item before and after
detection and filtering.