NEVAC-Cell-Phone

VAST 2008 Challenge
Mini Challenge 3: Cell Phone Calls

Authors and Affiliations:

Junyan Luo, The Pennsylvania State University, jzl106@psu.edu [PRIMARY contact]

Anthony Robinson, Penn State, arobinson@psu.edu

Chris Weaver, The Pennsylvania State University, cweaver@psu.edu

Student team: NO

Tool(s):

Improvise is a desktop application for building and browsing a wide range of flexible and powerful visual analysis tools. Live design of visual queries facilitates fast and flexible interactive drill-down into fine-grain relationships buried in spatiotemporal and social network information spread across multiple data sets. Cross-filtering queries across multiple views provides analysts with the means to seek out and dissect subtle patterns in complex information spaces.

We used Improvise to build cell.viz, an interactive visualization of the cell phone call data. The interface enables analysis of temporal patterns of phone calls, using two kinds of graph for multidimensional drill-down into social networks affected by geography. (Click the preceding two links above to get the visualization and instructions for running it.)

Two Page Summary: NO

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

PhoneNodes.txt

PhoneLinks.txt

---

Phone-2  Characterize the changes in the Catalano/Vidro social structure over the ten day period.

Detailed Answer:

1. Overview

Our toolkit provides multiple coordinated views to allow characterization of the Catalano / Vidro social structural over the ten day period.  Its main components (Figure 1) include:

- A map view that shows the cell tower locations on the island;

-A calendar view that shows the aggregated call time in each hour during the 10 day period.  Each row of the calendar contains 24 cells representing 24 hours, and their colors indicate the amount of aggregated call time.

-A graph view that shows the connections between callers.

-A slipstream view that indicate the sequences of phone calls.  It contains Bezier curves plotted across horizontal time line.  Each Bezier curve represents a phone call which starts from a time point in the upper line and ends with a time point in the lower line.

Activity timelines for towers, callers, and receivers.

Figure 1: Overview of the cell.viz.

Linking-and-brushing are enabled across the views.  For example, clicking on a caller in the graph view will lead the calendar view to show the hourly call time distribution of that particular caller. Similarly, clicking on a Bezier curve in the slipstream view will highlight the corresponding callers in the graph view and the hosting cell tower in the map.

Click here to download the video that demonstrates basic functionality of the tool.

2. Structure of the Catalano / Vidro Social Network

First we created a graph view that includes all phone call records for caller 200 (Figure 2). The call attribute graph view shows that caller 200 contacts seven unique callers: 1, 2, 3, 5, 97, 137.  If we explore the call time distribution for each of these callers, we find an interesting temporal pattern shared among 1, 2, 3, 5, and 200, where almost no calls are made in June 8th and very few calls are made after June 7th. Edge weights suggest caller 5 is the one 200 has called most frequently, and the node size (scaled according to degree centrality) indicates caller 1 is the most important among 1, 2, 3. Therefore we can assume 5 to be Estaban Catalano and 1 to be David Vidro. Callers 2 & 3 must then be Juan / Jorge Vidro - however, we do not have conclusive information to tell which is which.

Figure 2: Cell.viz filtered to show call activity for caller 200.

3. Temporal Changes

We can use the calendar view to filter other views through time (Figure 3). The caller network graph shows the call frequency between callers, and the call attribute graph shows callers (yellow is the call originator, red is the call receiver), the towers their calls received (green), and the date of interest (purple). Edges in the caller network graph indicate the duration of calls, while edges in the call attribute graph indicate the number of calls.

Figure 3: The cell.viz filtered to show call activity and relationships between callers on June 5^th.

Caller 1 made calls to 2, 3, and 5 at the same time on two occasions (June 1st & 5th), with 200 also on the line on the first occasion. This leads us to believe that caller 1 is organizing things and therefore is likely to be David Vidro.

We suspect that after initial preparations were made the plan was put into action and it became important for David Vidro to play a more active organizational role. Through June 6^th-7^th, more calls were directed toward 2 and 3. We see this as a sign of last minute preparations, where key members of the group were working hard to carry out the plan.

June 8th was a quiet day for the group - only caller 2 received a single call in the evening. This indicates to us that the plot was happening off the island. Records from June 9th and 10th show all members of the group began making phone calls again (Figure 4), but never to each other and their overall call time dropped sharply. This means they made it back home, and were likely notifying their families & friends or attempting to behave normally to avoid suspicion.

Figure 4: Call activity following the return of group members on June 9^th and 10^th.

Finally, we noted that the calendar view shows that several nodes (300, 306, 309, 360, 397) have call distribution patterns that developed after June 8th (Figure 5). These callers are well connected with caller 97.

Figure 5: Pattern after June 7th

4. Our Conclusions

Based on our call activity analysis with the cell.viz, we have the following hypotheses:

- After careful planning (as suggested by call durations from June 1^st-7^th), some sort of planned action was carried out by the Catalano / Vidro group on or around June 8th.  Given the absence of phone calls on June 8th and the geographical locations of 1, 2, 3, 5, we suspect they departed the island from near tower 29 & 30 and conducted the action off the island.

- Callers 97 and 137 are friends of the Catalano families, as they only have connections with 200 and 5.

- The group's activities triggered reaction among callers like 300, 306, 309, 360, and 397. Since those nodes had little call time before the event, and they were well connected with 97, we guess they were investigators and that caller 97 is either a prime suspect or an informant.