Table 3 Macro-F1 scores for temporal event reconstruction in empirical datasets. We highlight in bold the two best scores for each dataset. For baseline models we underline their highest score
From: Time-varying graph representation learning via higher-order skip-gram with negative sampling
Model | Operator | Dataset | ||||
|---|---|---|---|---|---|---|
LyonSchool | SFHH | LH10 | Thiers13 | InVS15 | ||
DyANE | Average | 56.4 ± 0.4 | 52.9 ± 0.5 | 52.3 ± 0.6 | 51.0 ± 0.4 | 52.7 ± 0.4 |
Hadamard | 89.7 ± 0.3 | \(\underline{86.5}\pm 0.3\) | \(\underline{74.6}\pm 0.6\) | 94.7 ± 0.1 | 94.1 ± 0.1 | |
Weighted-L1 | 90.2 ± 0.2 | 83.3 ± 0.5 | 73.3 ± 0.7 | 94.7 ± 0.1 | 94.4 ± 0.2 | |
Weighted-L2 | \(\underline{90.6}\pm 0.2\) | 84.5 ± 0.5 | 72.0 ± 0.5 | \(\underline{95.0}\pm 0.1\) | \(\underline{94.8}\pm 0.2\) | |
Concat | 65.7 ± 0.4 | 53.8 ± 0.4 | 56.2 ± 0.6 | 57.0 ± 0.4 | 50.9 ± 0.4 | |
DynGEM | Average | 57.7 ± 0.5 | 56.8 ± 0.7 | \(\underline{54.8}\pm 1.5\) | 40.4 ± 1.5 | 42.8 ± 0.9 |
Hadamard | \(\underline{62.2}\pm 0.4\) | 55.1 ± 1.0 | 52.5 ± 1.6 | 40.8 ± 1.5 | 43.7 ± 1.0 | |
Weighted-L1 | 58.4 ± 0.6 | 52.3 ± 0.7 | 50.9 ± 1.2 | \(\underline{41.3}\pm 1.6\) | 44.8 ± 0.9 | |
Weighted-L2 | 53.7 ± 0.6 | 47.0 ± 0.8 | 47.0 ± 1.3 | 39.2 ± 1.2 | 43.6 ± 0.6 | |
Concat | 60.4 ± 0.4 | \(\underline{57.8}\pm 0.3\) | 48.9 ± 1.7 | 36.9 ± 1.3 | \(\underline{45.7}\pm 1.0\) | |
DynamicTriad | Average | 51.7 ± 0.2 | 56.9 ± 0.4 | 60.2 ± 0.6 | 58.1 ± 0.2 | 56.1 ± 0.3 |
Hadamard | 60.3 ± 0.3 | 58.9 ± 0.4 | 59.5 ± 0.5 | 62.2 ± 0.3 | 64.7 ± 0.3 | |
Weighted-L1 | \(\underline{79.1}\pm 0.4\) | 72.3 ± 0.4 | 75.5 ± 0.6 | 70.8 ± 0.3 | 78.1 ± 0.2 | |
Weighted-L2 | 77.4 ± 0.4 | \(\underline{73.4}\pm 0.4\) | \(\underline{77.4}\pm 0.5\) | \(\underline{72.4}\pm 0.2\) | \(\underline{78.9}\pm 0.3\) | |
Concat | 52.2 ± 0.2 | 53.4 ± 0.3 | 55.9 ± 0.7 | 55.1 ± 0.2 | 53.2 ± 0.3 | |
DySAT | Average | 51.1 ± 0.3 | 49.6 ± 0.4 | 51.6 ± 0.5 | 50.4 ± 0.2 | 50.1 ± 0.3 |
Hadamard | \(\underline{75.1}\pm 0.5\) | \(\underline{52.9}\pm 0.3\) | 54.8 ± 0.6 | \(\underline{71.1}\pm 0.4\) | \(\underline{66.8}\pm 0.5\) | |
Weighted-L1 | 72.4 ± 0.5 | 51.5 ± 0.3 | 56.1 ± 0.6 | 66.4 ± 0.4 | 64.8 ± 0.3 | |
Weighted-L2 | 72.4 ± 0.5 | 51.7 ± 0.3 | \(\underline{56.8}\pm 0.7\) | 66.5 ± 0.4 | 63.7 ± 0.4 | |
Concat | 50.0 ± 0.3 | 50.1 ± 0.4 | 52.3 ± 0.5 | 49.8 ± 0.2 | 50.9 ± 0.3 | |
ISGNS | Average | 53.4 ± 0.4 | 50.3 ± 0.5 | 48.1 ± 0.6 | 49.4 ± 0.4 | 45.9 ± 0.5 |
Hadamard | \(\underline{90.1}\pm 0.3\) | 87.2 ± 0.4 | 80.8 ± 0.7 | 96.7 ± 0.2 | 96.7 ± 0.2 | |
Weighted-L1 | 89.9 ± 0.3 | 87.7 ± 0.4 | 81.6 ± 0.4 | 96.8 ± 0.2 | 96.4 ± 0.2 | |
Weighted-L2 | 89.7 ± 0.3 | \(\underline{\mathbf{88}\boldsymbol{.}\mathbf{2}}\pm \mathbf{0}\boldsymbol{.}\mathbf{4}\) | \(\underline{\mathbf{81}\boldsymbol{.}\mathbf{7}}\pm \mathbf{0}\boldsymbol{.}\mathbf{5}\) | \(\underline{\mathbf{96}\boldsymbol{.}\mathbf{9}}\pm \mathbf{0}\boldsymbol{.}\mathbf{1}\) | \(\underline{\mathbf{96}\boldsymbol{.}\mathbf{8}}\pm \mathbf{0}\boldsymbol{.}\mathbf{2}\) | |
Concat | 57.1 ± 0.5 | 50.2 ± 0.4 | 48.8 ± 0.7 | 52.7 ± 0.4 | 43.8 ± 0.4 | |
\(\text{HOSGNS} ^{(\text{stat})}\) | Hadamard | \(\mathbf{{98}\boldsymbol{.}\mathbf{5}}\pm \mathbf{0}\boldsymbol{.}\mathbf{1}\) | \(\mathbf{{98}\boldsymbol{.}\mathbf{8}}\pm \mathbf{0}\boldsymbol{.}\mathbf{1}\) | \(\mathbf{{99}\boldsymbol{.}\mathbf{8}}\pm \mathbf{0}\boldsymbol{.}\mathbf{1}\) | \(\mathbf{{99}\boldsymbol{.}\mathbf{6}}\pm \mathbf{0}\boldsymbol{.}\mathbf{1}\) | \(\mathbf{{99}\boldsymbol{.}\mathbf{1}}\pm \mathbf{0}\boldsymbol{.}\mathbf{1}\) |
\(\text{HOSGNS} ^{(\text{dyn})}\) | Hadamard | 90.3 ± 0.2 | 80.9 ± 0.4 | 68.1 ± 0.7 | 93.5 ± 0.2 | 87.2 ± 0.2 |
\(\text{HOSGNS} ^{(\text{stat}|\text{dyn})}\) | Hadamard | 91.8 ± 0.2 | 86.7 ± 0.4 | 73.6 ± 0.6 | 94.3 ± 0.1 | 89.0 ± 0.2 |