Skip to main content

Table 7 PubTator random-slice and time-slice results

From: Neural networks for link prediction in realistic biomedical graphs: a multi-dimensional evaluation of graph embedding-based approaches

   Random slice Time slice
  Node AUC AUC   Avg. Prec AUC AUC   Avg. Prec
Method combination (ROC) (PR) MAP R-prec @ k (ROC) (PR) MAP R-prec @ k
Deep- Average 98.85 99.01 83.67 75.97 99.93* 93.86* 95.51* 70.78* 62.16* 99.89
Walk Concat 99.20 99.30 91.01 85.46 99.94* 93.99 95.70 71.11 62.65 99.89
LINE Average 99.10* 99.23* 90.36* 84.56 99.97 88.68* 92.27* 55.61* 46.41* 99.89
  Concat 99.13 99.24 90.07 84.03 99.95* 90.32 93.01 62.51 53.21 99.89
node- Average 98.71 98.90 82.98 75.29 99.94* 88.40 92.07 55.72 46.48 99.87
2vec Concat 99.16 99.21 88.94 82.14 99.92* 88.13 91.83 53.24 43.69 99.84
AA N/A 92.92 84.56 56.48 66.38 83.33 85.10 80.24 35.49 40.13 90.56
CN N/A 98.40 98.28 79.84 87.10 99.94* 88.37 88.83 43.67 46.59 99.84
JI N/A 92.36 87.59 65.44 59.74 91.21 86.08 83.52 38.66 38.75 94.27
  1. (Bold: best score, *: not statistically different from best)