Skip to main content

Table 4 Average AUC scores for different autoencoder architectures. We report overall scores, and scores on networks with differing overlap density \(O_n\), for must-link and cannot-link sets separately

From: Overlapping community finding with noisy pairwise constraints

Architecture Small Networks Large Networks
Must-link constraints Cannot-link constraints Must-link constraints Cannot-link constraints
Overall \(O_n=10\)% \(O_n=50\)% Overall \(O_n=10\)% \(O_n=50\)% Overall On=10% \(O_n=50\)% Overall \(O_n=10\)% \(O_n=50\)%
AE1 0.634 0.651 0.616 0.779 0.820 0.739 0.456 0.452 0.460 0.844 0.906 0.782
AE2 0.625 0.652 0.597 0.785 0.828 0.742 0.442 0.443 0.442 0.849 0.906 0.791
AE3 0.624 0.667 0.580 0.796 0.843 0.748 0.422 0.416 0.429 0.849 0.905 0.794
AE1_l 0.659 0.705 0.614 0.737 0.777 0.697 0.453 0.445 0.461 0.789 0.842 0.735
AE2_l 0.657 0.718 0.597 0.776 0.812 0.739 0.470 0.444 0.496 0.780 0.847 0.714
AE3_l 0.654 0.773 0.534 0.835 0.877 0.793 0.424 0.407 0.440 0.879 0.919 0.839
All AE 0.642 0.694 0.590 0.785 0.826 0.743 0.444 0.434 0.45 0.832 0.887 0.775
\