Fig 1.
The general composition of an attack profile.
Table 1.
Features of attack models.
Fig 2.
Prediction shift in user-based collaborative filtering with attack size varies and filler size varies.
Fig 3.
Overall structure of the proposed shilling detecting method.
Fig 4.
Rating deviation distribution.
Table 2.
Attackers in suspicious rating segments ratio in phase 1 when attack size and confidence coefficient vary.
Table 3.
Attack detection ratio when attack size varies under confidence coefficient 90%.
Fig 5.
Detection rate and false positive rate when attack size varies.
Fig 6.
Comparisons of detection results with other methods.
Fig 7.
Detection results with different datasets with attack size varies.