Social Network Analysis of the Planetary Data System Final Presentation Kate Martin Mark Avnet May 9, 2006 ESD.342 Advanced System ArchitectureOverview Planetary Data System Image courtesy of NASA.Decomposing the PDS Database into 3 Bipartite Networks … PDS Database Data Set 1-Mode Data Set 2-Mode Instrument Host 2-Mode PDS Node 2-Mode Author 1-Mode UnweightedWeighted … Yields 12 separate 1-Mode networks for comparative analysisNewman-Type Table for the 12 PDS Networks • Network-Specific Results – Small-magnitude ǩvalues, but the regions that follow a power law are small for many of the networks – Short path lengths and high clustering coefficients – Degree Correlation • “Social” and “information” networks have r > 0 • “Technological” networks have r < 0 (except in one case where r << 1)Newman-Type Table for the 12 PDS Networks • Network-Specific Results – Small-magnitude ǩvalues, but the regions that follow a power law are small (and possibly non-existent) for many of the networks – Short path lengths and high clustering coefficients – Degree Correlation • “Social” and “information” networks have r > 0 • “Technological” networks have r < 0 (except in one case where r << 1)Newman-Type Table for the 12 PDS Networks • Network-Specific Results – Small-magnitude ǩvalues, but the regions that follow a power law are small for many of the networks – Short path lengths and high clustering coefficients – Degree Correlation • “Social” and “information” networks have r > 0 • “Technological” networks have r < 0 (except in one case where r << 1)Newman-Type Table for the 12 PDS Networks • Network-Specific Results – Small-magnitude ǩvalues, but the regions that follow a power law are small for many of the networks – Short path lengths and high clustering coefficients – Degree Correlation • “Social” and “information” networks have r > 0 • “Technological” networks have r < 0 (except in one case where r << 1)Affiliation networks lead to triangle motifs and high clustering coefficients Authors Data Sets Network of authors with data sets as edgesNewman-Type Table for the 12 PDS Networks • Network-Specific Results – Small-magnitude ǩvalues, but the regions that follow a power law are small for many of the networks – Short path lengths and high clustering coefficients – Degree Correlation • “Social” and “information” networks have r > 0 • “Technological” networks have r < 0 (except in one case where r << 1)Newman-Type Table for the 12 PDS Networks • Network-Specific Results –Small ǩvalues, but the regions that follow a power law are small for many of the networks – Short path lengths and high clustering coefficients – Degree Correlation • “Social” and “information” networks have r > 0 • “Technological” networks have r < 0 (except in one case where r << 1)Newman-Type Table for the 12 PDS Networks • General Observations – Results depend on the choice of how to represent the network – Weighting can give a more accurate picture but can also cause the metrics to lose their meaningNewman-Type Table for the 12 PDS Networks • General Observations – Results depend on the choice of how to represent the network – Weighting can give a more accurate picture but can also cause the metrics to lose their meaningNewman-Type Table for the 12 PDS Networks • General Observations – Results depend on the choice of how to represent the network – Weighting can give a more accurate picture but can also cause the metrics to lose their meaningCompletely Connected Areas ==> High clustering coefficients and High Pearson degree correlationr = 0.68r = 0.039# Nodes w/ k >=# Nodes w/ k >= Cumulative Degree Distributions 450 Authors by Instrument Host 120 Instrument Hosts400 100 350 # Nodes w/ k >= # Nodes w/ k >=300 80 250 60 200 150 40 100 20 50 0 0 0 20 40 60 80 100 120 140 160 0 5 10 15 20 25 30 35 40 45 50 Degree Degree 1000 100 10 100 10 1 Authors by Instrument Host Instrument Hosts 0 50 100 150 200 250 300 Degree 1 0 5 10 15 20 25 30 DegreeCommunity Structure Do PDS nodes represent communities? NO Node size: Betweenness Shape: Newman-Girvan Color: PDS Node AssociationCommunity Structure Do Instrument Hosts represent communities? YES Node size: Betweenness Shape: Newman-Girvan Color: Instrument Host AssociationCommunity Structure Do Instrument Hosts represent communities? YES Node size: Betweenness Shape: Newman-Girvan Color: Instrument Host AssociationCentrality and Network Representation • As before, weighting can affect the results. – Higher centrality without weighting. – Some measures are affected, and others are not. • Some measures are not valid for some networks. • Centrality tends to be highest for the network of PDS Nodes (with authors as edges).Centrality and Network Representation • As before, weighting can affect the results. – Higher centrality without weighting. – Some measures are affected, and others are not. • Some measures are not valid for some networks. • Centrality tends to be highest for the network of PDS Nodes (with authors as edges).Centrality and Network Representation • As before, weighting can affect the results. – Higher centrality without weighting. – Some measures are affected, and others are not. • Some measures are not valid for some networks. • Centrality tends to be highest for the network of PDS Nodes (with authors as edges).Centrality and Network Representation • As before, weighting can affect the results. – Small Bodies is a much better center without weighting. – Some measures are affected, and others are not. • Some measures are not valid for some networks. • Centrality tends to be highest for the network of PDS Nodes (with authors as edges).Centrality and Network Representation • As before, weighting can affect the results. – Higher centrality without weighting. – Some measures are affected, and others are not. • Some measures are not valid for some networks. • Centrality tends to be highest for the network of PDS Nodes (with authors as edges).What’s Your Szego Number? • Top two “best centers” determined for each of the 12 networks by each of the 4 metrics • The most commonly appearing nodes selected as the “best centers” for each network (above) – Not
View Full Document