Table 3. Comparison of algorithms for detecting link communities on some real networks.
| MDL-values | |||
| Datasets | Ball’s method with IB | NMFIB | Ahn’s method |
| Zachary’s karate club | 5.2072 | 5.2117 | 5.2200 |
| Dolphin social network | 5.4079 | 5.0291 | 6.1854 |
| High school friendship | 5.9890 | 5.5903 | 5.9488 |
| Les Miserables | 5.2863 | 5.1267 | 5.3196 |
| Political books | 6.8298 | 5.9248 | 7.1717 |
| Word adjacencies | 7.1179 | 6.3625 | 7.3043 |
| American college football | 7.0854 | 6.6730 | 7.1398 |
| Jazz musicians collaborations | 8.7322 | 7.6698 | 8.0360 |
| C. Elegans neural | 10.0340 | 8.4119 | 10.6335 |
| E. coli metabolic | 8.9949 | 8.7959 | 9.7428 |
| E-mail network URV | 10.4825 | 10.0901 | 11.7598 |
| Political blogs | 10.7068 | 9.4971 | 12.1782 |
| Network science collaborations | 4.2796 | 4.2834 | 4.1812 |
| Power grid | 10.5560 | 9.8559 | 8.9819 |
| Protein-protein interaction | 8.7585 | 8.6888 | 9.8867 |
| Word association | 12.8988 | 12.0587 | 14.5691 |
Here, Ball’s method and our NMF method both use the strategy of iterative bipartition (IB) to automatically determine the number of communities.