Table 1.
Network description and comparison.
The table lists six network metrics for seven different networks. The original disconnected graphs for meetings and phone calls are labelled as Meeting-DC (1st column) and Phone-Call-DC (2nd column), respectively. Their union forms the Unified-DC network (3rd column). After isolating the largest connected components, we obtain the Meeting Graph (4th column), Phone-Call Graph (5th column), and Unified Graph (6th column). The final column presents averaged statistics for 1,000 random graphs, each having the same number of nodes and edges as the Unified Graph.
Fig 1.
Different colours represent different clans within the mafia network. A red circle around a node signifies leadership roles within the network, including positions such as bosses and executives. Additionally, the size of a node is proportional to its degree, and the width of a link is proportional to its weight.
Table 2.
Classification of nodes based on family and leadership roles.
Fig 2.
A snapshot of a graph during the BTW-DEG disruption process.
After removing 28 nodes, the original graph has fragmented into cliques of different sizes: one 5-clique, four 3-cliques, and eight 2-cliques, with the remaining nodes being isolated. Since betweenness centrality can no longer discriminate between the nodes, degree centrality will be used to select a node from the 5-clique.
Table 3.
Dynamic disruption of various batch sizes, measured in R@20% and R@40%.
Nine disruption strategies are included, performed both on the entire graph and on the LCC (referred to as Standard and W-LCC, respectively). The lowest R-values for each column are put in bold, and the lowest values across two columns are asterisked.
Table 4.
Paired t-test results showing the differences between the standard and within LCC approaches for different values of batch size.
A significant p-value (p ≤ 0.05) indicates a statistically significant difference between the two approaches.
Table 5.
Comparison of best-performing strategies at each batch size b with the naive greedy disruption approach (GRD), measured in R@20% and R@40%.
The lowest R-values are highlighted in bold font.
Fig 3.
Structural characteristics of the nodes removed by GRD.
A large percentage of nodes, across different batch sizes, rank highly in the betweenness centrality and degree centrality, and are also articulation points.
Fig 4.
A detailed disruption process.
With the batch size setting to 3, 27 nodes are removed in 9 steps.
Table 6.
Performance comparison of the best performing structure-metric-based approach, GRD, and SF-GRD, measured in R@20% and R@40%.
The number of highest-ranking nodes t is set to be 5 in the experiment.
Table 7.
Detailed node categories using different disruption approaches, highlighting the first three steps for b = 3.
Nodes marked with ‘F’ belong to mafia families, while ‘L’ denotes a leadership role.
Table 8.
Time complexity and actual experimental time comparison.
N is the number of nodes in a graph, E is the number of edges, b is the batch size, and t′ is the size of the search space, set to be 15 in the experiment.