Figure 1.
A circle denotes a node, an arrow represents a trust relationship between two nodes, and the associated weight denotes the trust score.
Table 1.
Bias scores by the MB algorithm.
Table 2.
Bias scores by the -AVG algorithm.
Table 3.
Bias scores by the -MAX algorithm.
Table 4.
Prestige scores by different algorithms.
Table 5.
Summary of the datasets.
Table 6.
Comparison of bias by our algorithms and MB algorithm under AUC metric (top 5% nodes of the dataset).
Table 7.
Comparison of bias by our algorithms and MB algorithm under Kendall Tau metric.
Figure 2.
Comparison of prestige by our algorithms and MB algorithm in Kaitiaki dataset.
Three methods (AA, HITS, PageRank) are used as baselines for measuring the rank of prestige. The higher Kendall Tau value exhibits higher rank correlation between different algorithms and the baselines.
Figure 3.
Comparison of prestige by our algorithms and MB algorithm in signed trust networks.
Figure 4.
Robustness of bias by our algorithms and MB algorithm in Epinions dataset under (a) dishonest voting attack and (b) clique attack models.
The curves show the robustness of bias by our algorithms and MB algorithm at different spammer ratio. The larger Kendall Tau value implies that the algorithm is more robust. The robustness decreases as the spammer ratio increases. Note that the robustness of our algorithms are consistently better than the MB algorithm under both (a) dishonest voting attack and (b) clique attack models.
Figure 5.
Robustness of prestige by our algorithms and MB algorithm in Epinions dataset under (a) dishonest voting attack and (b) clique attack models.
The curves show the robustness of prestige by our algorithms and MB algorithm at different spammer ratio.
Figure 6.
Scalability of the proposed algorithms.
The curves show that the running time of our algorithms increases linearly as the number of nodes increases.
Figure 7.
(a) The curves show the bias by our algorithms at different . (b–d) The figures show the prestige (compared with different baselines) by our algorithms at different
parameter values.