Figure 1.
(a) The relation between and
in different methods.
The parameters are set as in IARR and
in IARR2. (b) the dependence of the Pearson correlation
on
in IARR and IARR2 methods. The results in this figures are averaged over 10 independent realizations. The error bars are the corresponding standard deviations.
Figure 2.
(a) and (b) the AUC and of different algorithms to random rating spamming.
(c) and (d) different algorithms to malicious push rating spamming. The results in this figure are averaged over 10 independent realizations. The error bars are the corresponding standard deviations.
Figure 3.
(a) and (b) the dependence of AUC and on
in IARR and IARR2 methods in the random rating attack case.
(c) and (d) the dependence of AUC and on
in IARR and IARR2 methods in the malicious rating attack case. The results in this figure are averaged over 10 independent realizations. The error bars are the corresponding standard deviations.
Table 1.
Some basic characteristics of the real data sets considered in this paper.
Figure 4.
(a), (c) and (e) are the distribution of of the CR, IARR and IARR2 algorithms in Movielens data, respectively.
(b), (d), (f) are the distribution of of the CR, IARR and IARR2 algorithms in Netflix data, respectively.
is set as
in both IARR and IARR2 algorithms.
Figure 5.
(a) and (b) are the frequency distribution of object degree in Movielens and Netflix, respectively.
(c) and (d) are the relation between frequency and
in movielens and Netflix, respectively.
Table 2.
AUC values of different algorithms for the real-data sets.