Temporal Visualization And Analysis Of Social Networks
Temporal Visualization and Analysis of Social Networks
Peter A. Gloor*, Rob Laubacher
MIT
{pgloor,rjl}@mit.edu
Yan Zhao, Scott B.C. Dynes
*Dartmouth
{yan.zhao,sdynes}@dartmouth.edu
Abstract
This paper describes a visual social browser for exploring the evolution of social networks over
time. We consider the exchange of e-mails between actors as an approximation of social ties.
Our system analyzes the dynamic progression of communication patterns of e-mail traffic within
groups of individuals. It combines a discrete visualization view, a continuous visualization view,
and an adjacency matrix view. The goal of our work is to develop a framework of visual
temporal communication patterns of different types of collaborative knowledge networks. As a
first application, our tool is used to analyze communication patterns and make recommendations
for improved productivity in innovation communities in an emerging management consulting
practice.
Contact:
Peter A. Gloor
Center for Coordination Science
MIT
Cambridge, MA 02142
Tel: 1-617-253-7018
Fax: 1-617-253-4424
Email: pgloor@mit.edu
Key Words: social network analysis, temporal visualization, e-mail mining, collaborative
knowledge networks, animation
Acknowledgements: The authors would like to thank Tom Allen, Tom Malone, Hans
Brechbuhl, M. Eric Johnson, and Fillia Makedon for their advice and support.
Temporal Visualization and Analysis of Social Networks
Peter A. Gloor, Rob Laubacher, Yan Zhao, Scott Dynes
In this paper we introduce a visual browser for the visualization and analysis of social links (relationships). Our
visual browser displays the progression of communication networks between individuals over time. Our goal is to
come up with an environment for the analysis of the dynamics of communication in social spaces, while respecting
individual privacy. While there has been substantial research dedicated to visualizing static social networks as
directed graphs or adjacency matrices [Johnson 2000, Smith & Fiore 2001, Tyler, Wilkinson & Huberman 2002,
Van Alstyne & Zhang 2003], little work has been done so far to visualize the evolution of social networks over time
[Holme, Edling & Lijeros 2003.
Other researchers have analyzed email communication flow to study the community structure [Danah, Potter &
Viegas 2003, Ebel, Mielsch & Bornholdt 2002, Girvan & Newman 2001], or automatically identifying communities,
solely based on the frequency of the message exchange between individuals, by partitioning the total graph
[Guimera et. al. 2002]. For visualization and analysis of those networks, there is a wealth of social network analysis
tools available such as pajek [Batagelj & Mrvar, 1998], ucinet [Borgatti, Everett & Freeman, 1992], or AGNI
[Varghese & Allen, 1993]. While pajek and ucinet also include the option to animate the graph, they do not allow
temporal visualization similar to our own social browser.
Further research has used email data to map communication patterns from the perspective of the individual.
This work typically creates representations of past messages that allow an individual to see the personal network
implied by their prior e-mail traffic. Other studies have analyzed virtual communities by applying social network
analysis methods and metrics [Ahuja & Carley, 1999,].
System Overview
We have implemented a flexible but scalable architecture. E-mail messages are processed locally in three steps.
In the first step, the e-mail messages and mailing lists are parsed and stored in decomposed format in a database on
the local machine. In the second step the database can be queried to select messages sent or received by a group in a
given time period. In the third step the selected communication flows can be represented in our visual browser using
our own netgraph and static and dynamic innomap views [Gloor et. al. 2003] or SNA visualization tools such as
pajek and ucinet. Our own social browser is particularly optimized to add a temporal dimension to the visualization
and manipulation of e-mail-based communication of a large number of actors. While it is straightforward to
visualize static social networks as directed graphs or adjacency matrices, little work has been done so far to visualize
the evolution of social networks over time.
For the visual placement of vertices and edges we use the Fruchterman-Reingold graph drawing algorithm
[Fruchterman & Reingold 1991] for force-directed placement, which is commonly used to visualize social networks.
This method compares a graph to a mechanical collection of electrically charged rings (the vertices) and connecting
springs (the edges). Every two vertices reject each other with a repulsive force, and adjacent vertices (connected by
an edge) are pulled together by an attractive force. Over a number of iterations, the forces modeled by the springs
are calculated and the nodes are moved in a bid to minimize the forces felt.
Social network graphs attempt to represent the strength of social ties between parties. In our algorithm, we treat
the exchanges of e-mail between actors as an approximation of social ties. In our visualization a communication
initiated by actor A to actor B is represented as a directed edge from A to B. The more interaction between actors A
and B occurs, the closer the two representing vertices will be placed. The most connected actors are placed in the
center of the graph.
Our system currently includes three views: temporal visualization in a discrete and a continuous way as well as
an adjacency matrix view. For the continuous temporal visualization we propose a new algorithm, called the sliding
time frame algorithm described in the next section. While the three views have some common features, they work
independently and give observers different information.
Continuous Temporal View
To visually analyze the evolution of communication patterns over time, we developed a dynamic visualization
algorithm where the layout of the graph is automatically recalculated every day, resulting in an interactive movie.
The simplistic approach would be, for any given day, to base the graph structure on the communications that
occurred during this day. However, this approach does not take into account communications that happened before
this day or after this day in a specific time-frame. For our dynamic visualization, we propose a new algorithm based
on the FR algorithm: the sliding time frame algorithm.
Figure 1. Sliding time frame algorithm in “with history” mode
The basic idea of the sliding time frame algorithm is to display active ties between actors in a sliding time frame
covering a flexibly selected interval of n days starting from the specific day the visualization is showing. The
window frame moves forward day by day, and new ties are subsequently added to the graph each day until the
desired width n of the sliding time frame is reached. This time frame window allows users to foresee the activities
happening inside the time frame after the current day. By default, all the old communication activities before the
current time frame window are included in the layout of the graph.
In figure 1 the time frame moves to day d as the animation progresses. Thus, day d is the current day that the
visualization is showing and the current time frame is [d, d+n]. All communications through day (d+n) are
calculated and displayed, and if a communication takes place before or on day d, it is active. In the “No history”
mode (figure 2), only the edges in the current window are included. The time frame moves to day d as the animation
goes. Thus, day d is the current day that the visualization is showing and the current time frame is [d, d+n]. Only
communications inside the current time frame are calculated and displayed, and only communications on day d are
considered active.
Figure 2. Sliding time frame algorithm in “no history” mode
To define the amount of “animated action” and animation speed we are using “keyframes” adjustable by the
user based on the number of new edges appearing in the visualization. The animation of the changing layout is
interpolated between those keyframes.
Application: Correlating Temporal Communication Patterns With Innovation
We are aiming to distinguish temporal communication patterns typical of different types of innovation
networks. Our hypothesis is that innovation networks are core/periphery structures [Borgatti & Everett, 1999] with
small world properties. They consist of a central cluster of people, the core team, forming a network with low
centrality, but high density. The external part consists of a network forming a ring around the core team. It has
comparatively low density, but high centrality, thanks to the central team. The actors in the outer ring have low
betweenness centrality, as they are only connected to core team members, but not among themselves. In the
experiment outlined below we investigated this hypothesis and made initial correlations between these patterns and
success of the individual efforts, given our knowledge of the outcome of these endeavors and their communication
patterns.
Our dataset consists of a one-year e-mail archive of a 200 people global consulting practice. As an
approximation of the ego network of the practice leader we are using his mailbox, similarly we obtained the mailbox
of the practice coordinator as an estimate of his ego network. We are taking these two mailboxes as an
approximation of the organizational memory of the consulting practice. We distinguish 15 communities through
messages grouped by the practice coordinator and the practice leader into separate mailfolders. The 15 communities
consist of 8 innovation teams developing new consulting service offerings, of the sales and marketing activities, of a
weekly brownbag that was also used to coordinate global activities of the practice, of the organization of a global
Webinar, of the development of the practice Web site, and of the team handling the collaboration with software
vendors. As a measure of the performance of the communities we are taking the quality of the community output as
judged by the practice leader.
2
1
4
3
Figure 3. Four Screen shots of movie of Webinar visualization (sliding time frame 30 days)
and evolution of group betweeness centrality
Figure 3 illustrates the progress over time of the communication activities around preparing and conducting a
Webinar, i.e. a Web based conference. The picture in the top left of figure 3 shows the structure of the team
preparing the Webinar. This team has high density and relatively low group betweeness centrality, as shown by the
first arrow in the center of the graph. The picture in the top right of figure 3 shows a screen shot of the
communication pattern during the first time the Webinar was conducted. The practice leader (blue dot) is sending
and receiving information in a star structure, the graph in the center as pointed out by the second arrow displays now
relatively high centrality. The third picture at the lower left displays the team preparing a rerun of the Webinar,
again communicating with relatively low centrality (third arrow). The final screen shot in the lower right of figure 3
displays the practice coordinator (red dot) rerunning the Webinar, communicating in a star structure with relatively
high centrality (fourth arrow) with his audience.
For our test dataset our temporal analysis conveys new insights which would have been much more expensive
to obtain with traditional means. Our tool offers a fast way to find periods of low and high centrality, and to identify
periods of high productivity and information dissemination. Nevertheless we needed other contextual cues to obtain
a full understanding of the activities, such as interviews with community members and a content analysis of the e-
mail messages.
Our continuing goals are to gain deeper insights into the correlation of the evolution of online group dynamics
with innovation, and developing a theory of member roles in innovation communities.
References
[Ahuja & Carley, 1999] Ahuja, M. & Carley, K. 1999, Network Structure in Virtual Organizations.
Organization Science, Vol. 10, No 6. 741-757.
[Batagelj & Mrvar, 1998] Batagelj, V. & Mrvar, A. 1998, Pajek - Program for Large Network Analysis.
Connections 21, 2, 47-57
[Borgatti, Everett & Freeman, 1992] Borgatti, S. Everett, M. & Freeman, L.C. 1992, UCINET IV, Version 1.0,
Columbia: Analytic Technologies
[Borgatti & Everett, 1999] Borgatti, S. & Everett, M. 1999, Models of core/periphery structures, Social
Networks 21 375-395
[Cummings & Cross, 2003] Cummings, J. & Cross, R. 2003, Structural properties of work groups and their
consequences for performance, Social Networks
[Danah, Potter & Viegas, 2003] Danah, B. Potter, J. & Viegas, F. 2003, Fragmentation of identity through
structural holes in email contacts. Proc. Sunbelt XII.
[Ebel, Mielsch & Bornholdt, 2002] Ebel, H. Mielsch, L. & Bornholdt, S. 2002, Scale-free topology of e-mail
networks. arXiv:cond-mat/0201476v2 12 Feb
[Fruchterman & Reingold, 1991] Fruchterman, T.M.J & Reingold, E.M. 1991, Graph drawing by force directed
placement. Software: Practice and Experience, 21(11)
[Girvan & Newman, 2001] Girvan, M. & Newman, M.E.J.2001, Community structure in social and biological
networks. arXiv:cond-mat/0112110v1, 7 Dec.
[Gloor et. al., 2003] Gloor, P. Laubacher, R. Dynes, S. & Zhao, Y. 2003, Visualization of Communication
Patterns in Collaborative Innovation Networks: Analysis of some W3C working groups. Proc. ACM CIKM
2003, New Orleans, Nov. 5-6, 2003
[Guimera et. al., 2002] Guimera, R., Danon, L., Diaz-Guilera, A. Giralt, F. & Arenas, A. 2002, Self-similar
community structure in organizations. ArXiv:cond-mat/0211498 v1, 22 Nov.
[Holme, Edling & Lijeros, 2003] Holme, P. Edling, C.R. & Lijeros, F. 2003, Structure and Time-Evolution of
an Internet Dating Community, arXiv cond-math.0210514, 15 Oct.
[Johnson 2000] Johnson, S. 2003, Who Loves Ya, Baby? Discover, Vol. 24 No. 4 (April)
http://www.discover.com/apr_03/feattech.html
[Smith & Fiore, 2001] Smith, M. & Fiore, A. 2001, Visualization components for persistent conversations,
ACM CHI.
[Tyler, Wilkinson & Huberman, 2002] Tyler, J. Wilkinson, D. & Huberman, B. 2002, “Email as Spectroscopy:
A u t o m a t e d
D i s c o v e r y
o f
C o m m u n i t y
S t r u c t u r e
w i t h i n
O r g a n i z a t i o n s ”
http://www.hpl.hp.com/shl/papers/email/index.html
[Van Alstyne & Zhang, 2003] Van Alstyne, M. & Zhang, J. 2003, EmailNet: A System for Mining Social
Influence & Network Topology in Communication. University of Michigan working paper
[Varghese & Allen, 1993] Varghese, G. & Allen, T. 1993, Relational Data in Organizational Settings: An
Introductory Note for Using AGNI and Netgraphs to Analyze Nodes, Relationships, Partitions and
Boundaries. Connections, Volume XVI, Number 1 & 2, Spring
