Identifying Important Vertices in Large Real World Networks

Advisor Information

Sanjukta Bhowmick

Location

UNO Criss Library, Room 232

Presentation Type

Oral Presentation

Start Date

6-3-2015 9:45 AM

End Date

6-3-2015 10:00 AM

Abstract

Centrality measures on networks can provide vital information such as the location of hubs in the network, which entities are critical for information flow, and which vertices are the most centrally located within the network. Generally, vertices that have the highest value (rank) for a specific centrality measure are the ones that are most important in the context of that measure. Using a technique known as group testing we show that it is possible to identify the highest ranked vertices for closeness and betweenness centrality in a significantly faster time than using the standard method of calculating all of the centrality values and sorting them to identify the highest. Furthermore, we apply a recursive method to remove the highest vertices identified at each step in order to find the next highest vertices thus identifying a larger number of top ranked vertices than if we were to use group testing only once. We also show how we can filter out unimportant vertices using group testing. A new algorithm for computing the betweenness centrality of a specific vertex (instead of all of the vertices, as is done by the Brandes algorithm) is also presented and implemented in the group testing technique. Experimental results on various real-world as well as synthetic networks are presented to show the effectiveness of our methods.

This document is currently not available here.

COinS
 
Mar 6th, 9:45 AM Mar 6th, 10:00 AM

Identifying Important Vertices in Large Real World Networks

UNO Criss Library, Room 232

Centrality measures on networks can provide vital information such as the location of hubs in the network, which entities are critical for information flow, and which vertices are the most centrally located within the network. Generally, vertices that have the highest value (rank) for a specific centrality measure are the ones that are most important in the context of that measure. Using a technique known as group testing we show that it is possible to identify the highest ranked vertices for closeness and betweenness centrality in a significantly faster time than using the standard method of calculating all of the centrality values and sorting them to identify the highest. Furthermore, we apply a recursive method to remove the highest vertices identified at each step in order to find the next highest vertices thus identifying a larger number of top ranked vertices than if we were to use group testing only once. We also show how we can filter out unimportant vertices using group testing. A new algorithm for computing the betweenness centrality of a specific vertex (instead of all of the vertices, as is done by the Brandes algorithm) is also presented and implemented in the group testing technique. Experimental results on various real-world as well as synthetic networks are presented to show the effectiveness of our methods.