Table 5 The ablation experiments of MPANN
From: Multi-probe attention neural network for COVID-19 semantic indexing
Model name | EBP (%) | EBR (%) | EBF (%) | MaP (%) | MaR (%) | MaF (%) | MiP (%) | MiR (%) | MiF (%) | |
|---|---|---|---|---|---|---|---|---|---|---|
MPANN | MPANN-Default | 87.41 | 63.52 | 71.20 | 97.03 | 50.44 | 55.02 | 88.62 | 62.78 | 73.49 |
- Context-Term Attention | 94.86 | 52.90 | 64.25 | 98.70 | 41.51 | 46.28 | 93.50 | 52.97 | 67.64 | |
- Journal-Term Attention | 90.66 | 60.78 | 70.28 | 98.36 | 47.63 | 52.58 | 91.55 | 59.71 | 72.28 | |
- Journal-Context Attention | 86.59 | 59.85 | 68.29 | 96.51 | 47.78 | 52.29 | 88.39 | 59.32 | 70.99 | |
- Journal-Topic Attention | 87.61 | 62.29 | 70.36 | 97.06 | 49.40 | 54.01 | 88.88 | 61.60 | 72.77 | |
- Context-Topic Attention | 84.91 | 62.47 | 69.64 | 96.02 | 50.40 | 54.87 | 86.57 | 62.05 | 72.29 | |
MPANN-Naive | 97.17 | 41.89 | 54.76 | 98.89 | 36.06 | 40.78 | 98.02 | 42.81 | 59.60 | |