VAST 2008 Challenge: Social network dynamics using cell phone call patterns Romain Bourqui ∗ LaBRI Frederic Gilbert † INRIA Bordeaux Umang Sharan ‡ INRIA Bordeaux Paolo Simonetto § LaBRI Faraz Zaidi ¶ LaBRI ABSTRACT This paper is a summary of the contest entry submitted to the VAST contest 2008 mini challenge 3. The primary task of the mini chal- lenge was to characterize the Catalno/Vidro social network based on the cell phone call data provided and to characterize the tem- poral changes in the social structure over the ten day period. This paper summarizes the data analysis performed on the synthetic data set provided, describes the visualization algorithms and tools em- ployed and the key observations from our analysis. We use Tulip [1] for exploring the data set. Tulip is a scalable and flexible framework for visualizing large graphs providing the user an easy platform for exploring and manipulating large networks. Keywords: Visual analytics,Social networks,Spatio-Temporal vi- sualization,VAST contest Index Terms: H.5.2 [Information Systems]: Information Inter- faces and Presentation—User Interfaces; 1 I NTRODUCTION The goal of the Mini Challenge 3: Cell Phone Calls (Social Net- work) is to identify the social network and the temporal dynamics associated with the evolving network structure over a period of 10 days based on a set of cell phone call records. The data set con- sists of information about 9834 calls between 400 cellphones over a 10 day period in June 2006 in the Isla Del Sueno. The records are expected to provide critical information about the Catalano so- cial network structure. Given the quantity and nature of the data involved, we decided to employ Tulip to help analyze the evolution of the call structure over time. Tulip facilitates easy visualization of large graphs—such as call graphs and location graphs in this case. The rest of the paper discusses the design of the Tulip and its role in the data analysis task followed by the conclusion. 2 THE TULIP FRAMEWORK The Tulip graph visualization framework provides an efficient and modular environment in which tools can be experimented for the purpose of information visualization research. Tulip enables the user to draw and display huge graphs, enable seamless navigation through different parts of the graph, employ different geometric operations and algorithms like subgraph extraction easily through plugins. One of the most challenging aspects of the mini challenge was the data analysis. Scripts were written to extract the caller and callee ids, time stamp, call duration and the cell tower location from the raw cell phone call logs. Thereafter, we imported the induced call graph and location graph in Tulip to facilitate better visual anal- ysis. We define a call graph G C =( V, E C ) as an undirected graph where nodes V correspond to caller ids and edges E C correspond to calls between two nodes in the network. Similarly, we define a location graph G L =( V, E L ) as an undirected graph on nodes V ∗ e-mail: bourqui@labri.fr † e-mail: frederic.pierre.gilbert@inria.fr ‡ e-mail: umangsh@gmail.com § e-mail: paolo.simonetto@labri.fr ¶ e-mail: faraz.zaidi@labri.fr where e =(u, v) ∈ E L iff nodes u and v used the same cell tower to make a call at least once. Further, we define a weighing function f : E L → N on the edge set E L where, f (e)= n and e(u, v) ∈ E L implies that nodes u and v used the same cell tower at least n times in the phone call logs. We also define a temporal sequence of snap- shots for the call graphs and location graphs to analyze the temporal dynamics at a finer granularity. G t C and G t L ∀ t ∈ {1,2,...,10} rep- resents the temporal snapshots of G C and G L respectively, one for each day. Figures 1 shows G C in the center and G 1 C and G 5 C in the right column in Tulip. Figure 1: Call graphs G C , G 1 C and G 5 C Similarly, Figure 2 shows G L , G 1 L and G 5 L in Tulip. Tulip pro- vides several views for the best visualization of the data set—a central view for visualizing the graph as a node-link structure with zooming and panning tools for quick navigation and a list view con- taining node and link attributes in the sidebar enabling quick filter- ing and attribute-based pruning. Figure 2: Location graphs G L , G 1 L and G 5 L