Skip to main content

Table 6 Macro-F1 scores in temporal event reconstruction and missing event prediction for synthetic datasets. We highlight in bold the best two scores for each dataset. For baseline model we underline their highest score

From: Time-varying graph representation learning via higher-order skip-gram with negative sampling

Model

Operator

Dataset

OpenABM-2k-100

OpenABM-5k-20

Reconstruction

Prediction

Reconstruction

Prediction

DyANE

Average

52.2 ± 0.1

51.7 ± 0.1

51.9 ± 0.1

51.9 ± 0.1

Hadamard

\(\underline{76.4}\pm 0.1\)

\(\underline{72.4}\pm 0.2\)

\(\underline{\mathbf{90}}\boldsymbol{.}\mathbf{5}\pm \mathbf{0}\boldsymbol{.}\mathbf{3}\)

\(\underline{77.8}\pm 0.2\)

Weighted-L1

70.3 ± 0.1

67.4 ± 0.2

78.2 ± 0.7

70.5 ± 0.3

Weighted-L2

70.3 ± 0.1

67.7 ± 0.1

78.8 ± 0.5

70.9 ± 0.3

Concat

53.8 ± 0.1

54.6 ± 0.1

52.5 ± 0.1

52.5 ± 0.2

\(\text{HOSGNS} ^{(\text{stat})}\)

Hadamard

91.1 ± 0.1

87.0 ± 0.1

98.7 ± 0.1

86.0 ± 0.1

\(\text{HOSGNS} ^{(\text{dyn})}\)

Hadamard

78.7 ± 0.1

79.8 ± 0.2

82.8 ± 0.3

82.4 ± 0.2