Table 3.
Comparing the performance of the proposed method with that of SVM classifiers using different sets of features [2], including (1) message-based, (2) user-based, (3) propagation-based, and (4) combined, as well as two peer methods—(5) Yang's [1] and (6) Sun's [8]. “R” represents rumors; “C” represents credible messages; “W. Avg” represents weighted average of rumors and credible messages
| No. | Class | Precision | Recall | F-rate | AUC | No. | Class | Precision | Recall | F-rate | AUC |
|---|---|---|---|---|---|---|---|---|---|---|---|
| (1) | R | 0.708 | 0.636 | 0.672 | 0.659 | (4) | R | 0.781 | 0.767 | 0.772 | 0.818 |
| C | 0.700 | 0.648 | 0.677 | 0.684 | C | 0.802 | 0.730 | 0.775 | 0.805 | ||
| W.Avg | 0.704 | 0.640 | 0.669 | 0.687 | W.Avg | 0.761 | 0.782 | 0.776 | 0.812 | ||
| (2) | R | 0.719 | 0.667 | 0.71 | 0.728 | (5) | R | 0.732 | 0.741 | 0.735 | 0.749 |
| C | 0.703 | 0.728 | 0.713 | 0.708 | C | 0.718 | 0.729 | 0.721 | 0.758 | ||
| W.Avg | 0.711 | 0.697 | 0.710 | 0.713 | W.Avg | 0.725 | 0.733 | 0.728 | 0.754 | ||
| (3) | R | 0.687 | 0.606 | 0.654 | 0.685 | (6) | R | 0.712 | 0.683 | 0.703 | 0.701 |
| C | 0.698 | 0.769 | 0.738 | 0.742 | C | 0.677 | 0.679 | 0.677 | 0.705 | ||
| W.Avg | 0.691 | 0.718 | 0.702 | 0.718 | W.Avg | 0.705 | 0.680 | 0.688 | 0.702 | ||
| (7) | R | 0.829 | 0.803 | 0.813 | 0.839 | ||||||
| C | 0.797 | 0.789 | 0.791 | 0.828 | |||||||
| W.Avg | 0.812 | 0.793 | 0.799 | 0.831 | |||||||