Figure 2. Comparison of relative precisions of methods under different fractions of training sets on four real networks.

We compared relative precisions of eight methods under different fractions of training sets on the four networks and the precisions are returned with the average over 100 runs. The fraction of training sets f is varied from 0.3 to 0.9. The four networks are C. elegans, Jazz, USAir and Email. The link prediction methods are NMF1, SPM, NMF − D1, NMF − A1, CRA, NMF2, NMF − D2 and NMF − A2.