University College Dublin - SocialDynamicsVis
VAST 2008
Challenge
Mini Challenge 3: Cell
Phone Calls
Authors and Affiliations:
Michael Farrugia, University College Dublin, mike.farrugia@gmail.com
Aaron Quigley, University College Dublin, aquigley@ucd.ie [Faculty advisor]
Student team: YES
Tool(s):
A tool to visualize the Catalano family network was
specifically developed using Processing
during the time period allocated for the competition. The tool was developed to
enable easy animation of the network, and to visualize geographic movement of
connected people over time. When animating a network, the edges between nodes
increase width and colour intensity, as they get stronger, then start fading as
time passes. Nodes are energised using a spring based model, supplemented by
interaction features to allow manual movement of nodes. Each node has a context
menu to allow interaction and exploration. The tool provided a working platform
for visually analysing the network, and an opportunity to try visualization
techniques applied to dynamic social networks. We plan to extend the tool to
make it more generic, to support different time and geography based datasets.
Pajek was
used to calculate social analytic measures on the network. It was primarily
used to get an overall picture of the whole network structure, and to classify
the importance of people in the network based on different centrality measures.
Microsoft Excel was used for analysing attributes of the whole network
and individual actors in the network.
Two Page Summary: YES
Video:
ANSWERS:
Phone-1: What is the Catalano/Vidro social
network, as reflected in the cell phone call data, at the end of the time
period
Phone-2 Characterize the changes in
the Catalano/Vidro social structure over the ten day period.
Detailed Answer:
We started our analysis by developing a node-link
graph exploration visualization that allowed us to explore relationships in the
network and animate calls between them over time. Knowing that node 200 is
presumed to be Ferdinando Catalano we started looking at his immediate
neighbours and noticed that he communicates with a small group of people, who
are main coordinators of the network. This hierarchical network structure and
the overall cohesiveness of the network is representative of cults and other
criminal organizations run by families.
When animating the network of main coordinators and
their neighbours, we originally didn't notice any major changes in time. Later
we realised this was due to the call decay functions reducing the edge size too
slowly. We then tried to find a way of displaying the whole network, without
the clutter of the node link diagram. For this, we created the animated matrix
view, where each call is represented by a small square in a 400x400 grid.
When we animated the matrix view, we immediately saw a
high concentration of calls, shown as red dots, appearing in the 300's range on
day 8 (Figure
1 + Video). We then zoomed on these nodes from the
matrix display to the node-link visualization. In the node-link visualization
we saw that on this day, the new set of nodes (306,309,360,397) start
communicating with the same group of nodes that 1,2,3,5 communicated with
before. When we animated again with faster decay functions, we saw that the old
coordinating nodes stopped communicating when the new ones started calling.
Figure 1 - Matrix View up to day 9 showing
(1) increase of red dots around 300 range and (2) fading of old coordinator's
connections (click
on images for a high-res version)
When we animated the node-link visualization and let
the new nodes move freely towards people they were calling, they moved near the
original coordinating nodes (Figures 2 and 3 + Video). This meant that they
were calling the same people that the old nodes had been calling before. With
the help of this visualization, we found an exact 1-1 mapping between the old
and new coordinators. This was also confirmed using structural equivalence
measures in Pajek.
Figure 2 - Communication of coordinators
with their neighbours up to day 6. Colour = location; size = node importance
Figure 3 - Communication of coordinators
with neighbours between days 8 and 10. Colour = location; size = node
importance
Once we identified the groups of people who occupied
key roles in the family structure, we required a way to query the call records
of the individual people in the network. For this we developed a graphical and
data report for each cell phone in Excel (Figure
4). This report was integrated with the visualization
by selecting view information in the
context menu of a node in the node-link view.
Figure 4 - Profile card for each cell phone
with call data
Once we had a good indication of what happened on day
8, we started looking at possible reasons for this change and started testing
various hypotheses. This stage of hypothesis testing and exploration guided the
further development of the visualization tool. For instance, we added
geographic animation of nodes to explore geographic mobility of people meeting in
the same location, at the same time. Nodes are placed on an underlying map and
move around the map as they change location. We also added node colouring and
different node shapes based on node importance, to explore network hierarchy.
Some features implemented however, were not helpful in
providing insight. For instance, as a first attempt at examining location
change in all nodes, we developed an animated grid view of all nodes that
changed colour when the node changed location. Instead, we grouped all cell
towers in 5 regions, and created the graph in Figure
6. This graph shows periods of call activity including
a significant decrease in calls from sea during weekends. This suggests that
people may be earning their livelihood at sea.
The hypothesis testing stage was the most time
consuming phase of the process, since we found both supporting and
contradicting evidence to most hypotheses. However we are satisfied with the
way hypothesis testing guided the development of the visualizations. Each
hypothesis provided the questions that each visualization feature had to
answer, and the answers gave a rough measure on how effective the visualization
was.
Figure 5 - Isolation of Node 200
Scenario Hypothesis
There is strong contradicting evidence against the
hypothesis that a new group of coordinators were planning a coup against the
old coordinators. The fact that the people who communicate with the
coordinators immediately start
communicating with the new ones, with the same
pattern, suggests that the activity was pre-planned (Figure 3). The organisation of the changeover is also apparent
on the night/morning of days 7 and 8. The old controlling numbers communicate
with people at night (represented by squares), who communicate with the new
numbers early in the morning (Figure 7). The direction of the calls indicates
that these people knew about the new numbers because they initiated the call.
Furthermore the old coordinators don't attempt to regain their position on day
10, when they make phone calls again. The lack of peaks in activity before and
after the event (Figure
6) also does not support the hypothesis that something
hostile happened. Alternatively, if there was a peaceful shift of control, it's
difficult to explain why nobody calls the old coordinators on day 10 when they
start making calls again.
The evidence found supports the hypothesis that, cell
phone 200 is owned by the interim leader in absence of Ferdinando Catalano, who
is overseas until day 8. We know that David Vidrio coordinates high level
Paraiso activities, which suggests that he's most likely to be in charge of the
organisation in Ferdinando's absence, making him node 200. Due to a dispute,
Catalano decides to isolate David Vidrio on his return. From overseas, Catalano
instructs the other coordinators to replace their old phones and start using
new ones from day 8 (Figure 5). Since the old coordinators and the new ones
communicate with the exact same neighbours, and the coordinating groups never
communicate together, we believe they are the same people. Catalano is node 300
because he is the only node connected to all the coordinators after day 8. Node
300 communicates mostly with node 268 making him Esteban Catalano. The only two
subjects that remain loyal to David Vidrio, are his two brothers Jorge and Juan
Vidrio, nodes 137 and 97, who communicate with him after he is isolated.
However, there is no further information to distinguish between the two.
Figure 6 - Call activity by location and average duration of calls in minutes
Figure 7 - Changeover on night day 7 and morning
day 8