Fig 1.
Diagram describing processing steps of EdgeBoost.
Fig 2.
NMI of Baseline Community Detection Methods.
NMI of six community detection algorithms with varying percentages of removed edges δ. Error bars are not included because standard error is too small.
Fig 3.
RE of Baseline Community Detection Methods.
RE of six community detection algorithms with varying percentages of removed edges δ. Error bars are not included because standard error is too small.
Fig 4.
Intra-edge Precision of Link Prediction.
Precision plots of three link prediction algorithms: Adamic-Adar (left), Common Neighbors (middle), and Jaccard (Right) for various values of mixing parameter μ: 0.1 (top), 0.3 (middle), and 0.5 (bottom). The X-axis corresponds to number of top-k edges as scored by the link prediction algorithm as a percentage of the number of edges in the network. Intra-edge precision is on the y-axis.
Fig 5.
Histogram of edge weights on a benchmark graph with μ = 0.4 and 20% of the edges removed: scores from AA link predictor (top) and weights of co-community network (bottom).
Fig 6.
Karate Club Co-community Network.
Visualization of the co-community network for “Zachary’s karate club” network. Each panel shows the network pruned at various thresholds τ.
Fig 7.
EdgeBoost Performance on LFR Networks.
Performance of six popular community detection algorithms on the LFR benchmark networks. Dashed yellow bar shows the improvement of EdgeBoost over using the baseline community detection method.
Fig 8.
Performance of EdgeBoost (solid) and the baseline Louvain algorithm (dashed) on LFR benchmarks. The purple shaded region shows the improvement of EdgeBoost for NMI. The bottom row shows the relative error of the partition size.
Fig 9.
Performance of EdgeBoost on Standard Network Datasets.
Comparison of EdgeBoost on set of standard real network benchmarks community detection.
Fig 10.
Performance of EdgeBoost on Facebook Networks.
Comparison of EdgeBoost on ego-networks from Facebook.
Fig 11.
Varying NumIterations for EdgeBoost with Louvain.
The parameters are set as follows: μ = 0.2 (left) and μ = 0.5 (right) over δ values ranging from 0.0 to 0.6.
Fig 12.
Varying NumIterations for EdgeBoost with InfoMap.
The parameters are set as follows: μ = 0.2 (left) and μ = 0.5 (right) over δ values ranging from 0.0 to 0.6.
Fig 13.
Varying τ for EdgeBoost with Louvain.
Varying the co-community threshold (τ) for EdgeBoost with μ = 0.2 (left) and μ = 0.5 (right) over δ values ranging from 0.0 to 0.6.
Fig 14.
Varying τ for EdgeBoost with InfoMap.
Varying the co-community threshold (τ) for EdgeBoost with μ = 0.2 (left) and μ = 0.5 (right) over δ values ranging from 0.0 to 0.6.
Fig 15.
Comparison of the runtime between EdgeBoost and baseline Louvain algorithm on networks ranging from size 1000 to 128000 nodes. EdgeBoost has the NumIterations set to 50.