- Patrick Kenis (Free University of Amsterdam)
- Analysing Social Network Data by Means of Visualisation Techniques
- Abstract:
In this presentation it will be shown that social network data can
effectively be analysed by means of visualisation techniques. Usually
social network data are analysed by calculating structural and locational
property measures and/or by confronting them with specific theories. In
these contexts visualisations are often used to "illustrate" the findings.
This presentation will show that visualisation techniques can be used as an
additional way to analyse data. It will be argued and illustrated that
network visualisation can go far beyond "illustration". Network
visualisation can help to improve communication about the data to third
parties; it can help researchers to better explore specific properties of
certain networks or facilitate the exploration of differences across social
networks; or it could even serve to discover explanations for social
phenomena.
In order to reach the above aim, visualisation techniques are necessary
which go beyond those, which are available at the moment. Visualisation
techniques will be presented which are developed at the moment at the
University of Konstanz (at the Faculty of Mathematics and Computer Science
and the Faculty of Public Policy and Management). The techniques developed
are based on the principle that an effective visualisation is a combination
of providing an algorithmic solution to a substantive problem in such a way
that basic design principles are respected
The contention that effective visualisation can be an important instrument
in the analysis of social network data will then be illustrated by data
from a comparative social network study. The study compares 9 German
municipalities in their effectiveness for providing HIV preventive measures
for intravenous drug-users. The hypothesis being that it is the structural
properties of the drug policy networks in these municipalities, which
explain the degree of presence of preventive measures.
- Arthur L. Dryver, Martina Morris, John Potterat, and Stephen Muth (Penn State University)
- Visualizing Big Networks: The 8000 Nodes and 36000 Links in the Colorado Springs Study
- Abstract:
Visualization tools are an important component of both exploratory
and confirmatory data analysis. Multivariate data, and networks are
a form of this, present unique challenges for visualization. A good
example of the strengths and weaknesses of such tools can be seen
in Chernoff's faces. While these provide a clever visual analog for
multivariate clustering techniques, with much more intuitive access
to the underlying data structure, their usefulness is also limited to
small data sets. Network visualization tools often have similar
strengths and weaknesses. In the process of analyzing the Colorado
Springs Project 90 Network data set we have had to grapple with this
problem. The network data were collected over the span of five years
and contain information from 595 respondents on 8166 unique contacts,
and 36838 dyads over time. In this talk we will present some of the
successes and failures of our effort to visualize this network using
Pajek.
The application will focus on the use of network images in
the analysis of the overlay of sex and needle sharing networks.
- Douglas R. White (University of California at Irvine), Vladimir Batagelj, and Andrej Mrvar (University of Ljubljana)
- Analysis of Genealogies Using Pajek
- Abstract:
Pajek is a program package for analysis and visualization of
large graphs and networks. Genealogies are examples of large
networks available already in a computerized form on the
Internet.
Pajek supports, besides usual Ore-graph also p-graph
representation of genealogies, which is more convenient for their
visualization and analysis.
Several standard network analysis procedures can be used for
analysis of (large) genealogies represented as p-graphs, e. g.:
biconnected components, pattern searching, and different
statistics, including the relinking index as a measure of degree
of marital relinking among families.
Approaches to the creation of structural variables, the
decomposition of genealogies according to biconnected components,
lineages, etc., and the dynamic visualisation of genealogies
(kinemages) will also be presented.
- Ulrik Brandes (University of Konstanz)
- Centrality and Prestige Made Visible
- Abstract:
Effective network visualization rests upon two pillars:
clarity and substance. It is argued that current visualization
techniques typically focus on either one of these aspects.
A potential step forward is the integration
of specific network information directly into
a graphical design without neglecting clarity.
We thus derive new visualization techniques that aim at
exact representation of important structural variables
in nevertheless readable pictures. The two variables considered
here are centrality and prestige, and the images presented
combine the benefits of precise tabulation with the
convenience of graphical presentation. Interestingly,
precursors of these techniques can be found in the
not-so-recent literature (Northway 1940, Whyte 1943).
- Lothar Krempel (Max Planck Institute for the Study of Societies, Cologne)
- Visualising Networks with Spring Embedders: Properties and Extensions for Two-Mode and Valued Graphs
- Abstract:
This paper explores how the general ideas of a spring embedder can
be extended to treat graphs where the nodes are linked by forces of
different size (valued graphs) and graphs in which two distinct sets
of nodes are connected (two mode data).
Starting with a short account of the basic components of a spring embedder
(attractive and repulsive forces), we summarize modifications and
extensions of the basic concept reported in the literature. A closer
look at the behavior of the single components will give us a basic
understanding of the workings of these algorithms. The growing number
of force and field conceptualizations that can be used as ordering
principles point to future developments. This growing tool chest of
ordering principles waits to be exploited for the quasi-experimental
study of structures.
In a third part of this paper we demonstrate and give results on the
precision for implementations of spring embedders which are applied
to valued graphs (where relations have different strength) and two
mode information (relations between elements of two different sets).
- Linton C. Freeman (University of California at Irvine)
- Mapping a Three-Dimensional Structure onto a Two-Dimensional Plane
- Abstract:
Data on the airline distances linking the capitols of the
26 largest trading nations were used. Those nations were projected onto a
two dimensional plane using various standard algorithms from network
analysis. Results are evaluated in two ways:
- by correlating the 2D projected distances with the original 3D global distances, and
- by using the judgments of accuracy made by a collection of college students.
Comparing these results casts some light on issues of data reduction and
on questions about what people "see" in simple presentations based on
proximities/distances.
- Vladimir Batagelj, Matjaz Zaversnik, and Andrej Mrvar (University of Ljubljana)
- Partitioning Approach to Visualization of Large Networks
- Abstract:
Some approaches to partitioning in large relational networks
will be presented, like
- connectivity based partitions:
- standard concepts from graph theory (components, cliques, k-cores, ...)
- neighbourhoods of central vertices
- hierarchy of similar graphs
- neighbourhood based partitions:
- cluster is a set of units with similar neighbourhoods
(degree partition, regular partition, ...)
The obtained partitions are used for visualisation of given network
(analysis of main core and residual graphs, shrinking the main core,
deleting the main core, reordering the relational matrix).
Some illustrative examples will also be presented.
- Noshir S. Contractor (University of Illinois at Urbana Champaign)
- IKNOW: A Visualization Tool to Assist and Study the Emergence of Social and Knowledge Networks
- Abstract:
Because information transacted over electronic media such as the Web can be
stored in digital form, a new generation of software called `collaborative
filters' or `communityware' can be used to visualize a community's social and
knowledge structures. One such tool, IKNOW
(Inquiring Knowledge Networks On the Web),
has been designed by a team of UIUC
researchers to assist individuals to search the community's databases to
automatically visualize and answer questions about the community's knowledge
network, that is, Who knows what? as well as questions about the community's
cognitive knowledge networks, that is, Who knows who knows what? within the
community. Unlike traditional web search engines that help an individual
search for content on the web, tools such as IKNOW search for, and visualize,
content and contacts (direct and indirect). In addition to being instantly
beneficial to users, they also provide the researcher with an opportunity to
unobtrusively study and visualize the influence of these communityware tools on
the social structure within communities. This paper explores the rationale
behind the design of one communityware tool, IKNOW. Next it reports on how
several work and social communities are using IKNOW.
- Gerald R. Falkowski and Sheri L. Feinzig (IBM Corporation)
- Using ONA to Improve Teaming and Communications in IBM
- Abstract:
This paper discusses how we used visualization tools to conduct an
Organization Network Analysis (ONA) to improve teaming and
communications within IBM's "Fulfillment" operation, a sprawling
enterprise that employs 30,000 people in 122 countries, and handles the
paperwork and shipping logistics for some 14 million customer orders
yearly. This project began in 1996 in an attempt to see if network
visualization techniques could facilitate an organization's ability to
diagnose problems and make subsequent changes to address those
problems. To start, we used ONA to benchmark the best practices of 15
companies that had also undergone a similar business transformation.
This best practices study focused on identifying those things that were
integral parts of these companies' business transformation efforts.
Our analysis suggested that IBM Fulfillment ranked in the bottom third
of the studied companies in terms of overall organizational
effectiveness.
By mapping Worldwide Fulfillment's work interactions and information
exchanges both within the department and with the department's key
constituents within other parts of IBM, we discovered that: 1) nearly a
quarter of Worldwide Fulfillment employees were not connected to or
communicating with their colleagues; 2) a third had little or no input
on decision making; and 3) nearly half (44%) had no opportunity to
provide input into new ideas. Based on our findings, we facilitated a
series of workshops designed to use ONA visualization techniques to
gain insights and then developed an action plan.
During 1997 we conducted a second Organization Network Analysis to
gauge progress based on actions taken and found considerable
improvement. When compared to the 15 international benchmark
organizations, the IBM Fulfillment organization improved in rank to the
top 1/3 in terms of effective team communication that is
non-hierarchical, extensively networked, and frequent. Moreover, this
study demonstrated the ability of these techniques to highlight ways to
change organizations. The resulting ONA maps pointed out exactly which
teams and individuals needed to increase their communication, who
needed more input on decisions, and who needed to become more involved
in innovation. By using network visualization tools to identify gaps
and disconnects in these areas, Worldwide Fulfillment was able to
successfully address those issues.
