Accession Number : ADA600205


Title :   Incremental Centrality Algorithms for Dynamic Network Analysis


Descriptive Note : Doctoral thesis


Corporate Author : CARNEGIE-MELLON UNIV PITTSBURGH PA DEPT OF ELECTRICAL AND COMPUTER ENGINEERING


Personal Author(s) : Kas, Miray


Full Text : https://apps.dtic.mil/dtic/tr/fulltext/u2/a600205.pdf


Report Date : Aug 2013


Pagination or Media Count : 341


Abstract : The increasing availability of online resources such as digital libraries and social networking websites has led to an upsurge of interest in the analysis of social networks. To date, a wealth of social centrality measures have been designed for determining the importance of nodes in a social network from different aspects. A significant number of social centrality metrics depend on the shortest paths in the network usually requiring solving the all-pairs shortest path problem. However, most of these metrics were designed for static snapshots of 20-30 node networks. Computing centrality metrics in dynamically changing, large networks is almost unfeasibly costly, especially if it involves repeatedly calculating centralities from scratch for each incremental change. This thesis proposes incremental algorithms for the two of the most popular shortest-path based social centrality metrics \201e.g. closeness centrality and betweenness centrality\202 to avoid computations from scratch by performing early-pruning and achieving substantial performance improvements in dynamically changing networks. It explores the computational time savings and the memory requirements as the realistic social networks being analyzed scale to very large sizes. The key idea is to start with the old output of the algorithm and to modify/update only the affected values such that the changes in the network \201e.g. edge/node insertions/deletions/modifications\202 are reflected in the centrality values as well. The approximate versions of incremental closeness and betweenness centralities through k-hop bounded computations are also designed where the shortest paths included in the computations should be less than or equal to k forcing the centrality computations to remain within a k-hop subgraph of a node instead of the entire graph. The performance results obtained via experiments on a wide variety of synthetic and real-life dynamic networks suggest substantial improvements over the state of the ar


Descriptors :   *NETWORK ANALYSIS(MANAGEMENT) , ALGORITHMS , THESES


Subject Categories : Administration and Management


Distribution Statement : APPROVED FOR PUBLIC RELEASE