Extracting Skeleton of the Global Terrorism Network Based on m-Modified Topology Potential

Abstract

Skeleton network extraction is a crucial context in studying the core structure and essential information on complex networks. The objective of this paper is to introduce the novel network extraction method, namely, TPKS-skeleton, for investigating the global terrorism network. Our method aims to reduce the network’s size while preserving key topology and spatial features. A TPKS-skeleton comprises three steps: node evaluation, similarity-based clustering, and skeleton network reconstruction. The importance of skeleton nodes is quantified by the improved topology potential algorithm. Similarity-based clustering is then integrated to allow detecting high incident concentrations and allocating the important nodes according to the event features and spatial distribution. Finally, the skeleton network can be reconstructed by aggregating high-influential nodes from each cluster and their simplified edges. To verify the efficiency of the proposed method, we carry out three classes of a network assessment framework: node-equivalence assessment, network-equivalence assessment, and spatial information assessment. For each class, various assessment indexes were performed using the original network as a benchmark. The results verify that our proposed TPKS-skeleton outperforms other competitive methods in particular node-equivalence by Spearman rank correlation and high network structural-equivalence defined by quadratic assignment procedure. In the spatial perspective, the TPKS-skeleton network preserves reasonably all kinds of spatial information. Our study paves the way to extract the optimal skeleton of the global terrorism network, which might be beneficial for counterterrorism and network analysis in wider areas.

Motif-h: a novel functional backbone extraction for directed networks

“…Intuitively, nodes with a higher centrality value will be prior preserved as the core structure. With this respect, numerous researches and excellent works have proposed to handle such problem [46,51], which can be classified into three categories: coarse-graining model, filtering model and generalist model.…”