=Paper=
{{Paper
|id=Vol-443/paper-11
|storemode=property
|title=Interactively Visualizing Dynamic Social Networks with DySoN
|pdfUrl=https://ceur-ws.org/Vol-443/paper11.pdf
|volume=Vol-443
}}
==Interactively Visualizing Dynamic Social Networks with DySoN==
https://ceur-ws.org/Vol-443/paper11.pdf
Interactively Visualizing Dynamic Social Networks with
DySoN
Georg Groh Holger Hanstein Wolfgang Wörndl
TU München, TU München, TU München,
Faculty for Informatics, Faculty for Informatics, Faculty for Informatics,
Germany Germany Germany
grohg@in.tum.de holger@hanstein.net woerndl@in.tum.de
ABSTRACT as the dynamics of relation-type or relation-intensity) is a
The dynamic social network visualizer “DySoN” (Dynamic time-averaged view of a faster underlying relational dynam-
Social Networks) aims at understanding patterns and struc- ics. For example we might view the intensity of a friendship
tural changes in dynamic social networks that evolve over relation which is instantiated in time and space only four
time via an interactive visualization approach. times in the last month as less strong in that month com-
As an alternative and supplementation to the numerous other pared to a month where the same friendship was instantiated
approaches to visualization of social network data and as an eight times. With a more fine granular time view (e.g. weeks
attempt to overcome some of the drawbacks of these ap- or days), the four times of seeing your friend may be concen-
proaches, DySoN interactively visualizes streaming event trated in a short time interval, e.g. two days, while the rest
data of social interactions by an interactive three-dimensional of the month there was no interaction at all, thus yielding
model of interpolated NURBS ”tubes”, representing activ- high intensity in these two days and very low intensity for
ity and social proximity within a given set of actors during the other 28 days of the month. Thus the different underly-
a given time period by using three dimensions of temporal ing time granularities may give rise to the notions of social
information mapping: spatial density (tube distance), tube- micro- and macro-dynamics and social micro- and macro-
color and tube-diameter. contexts respectively. In that view the social macro-context
We use a self assembled large collaboration network of Jazz dynamics is determined by or at least strongly related to the
musicians with a straightforward semantics for the computa- the underlying social micro-context(s).
tion of relation strengths for the evaluation of the approach.
We also discuss applications of the concept for awareness With respect to a rather small time and space granularity,
services in mobile peer to peer social networks, which ex- everybody has an idea of such social micro-contexts or dy-
hibit a vivid measurable social micro dynamics in time and namic social structures, at least unconsciously, because some
space. of the mechanisms that organize mankind into groups and
hierarchies can be observed in real life when people form
Author Keywords changing patterns with their bodies in time and space while
Mobile Dynamic Social Networks, Awareness Services interacting socially. In order to get an abstract view of re-
lationships and their instantiations between actors in a so-
cial network, one can build upon these ”physical patterns”
ACM Classification Keywords
which may be measured and modeled mathematically, es-
H.5.2 Information Interfaces and Presentation: Miscellane- pecially when considering mobile interaction schemes with
ous—Optional sub-category community- or social network platforms [32]. Besides such
physical expressions of social relations a wealth of other
MOTIVATION & OBJECTIVES highly dynamical features or indicators of social relations
Social relations and social structures respectively are clearly exist that may be modeled (see section ).
of a dynamical nature. Many dynamical aspects of social
relations investigated by social science are rather long term However dynamical the social network model or social con-
which is partly due to the properties of the investigation in- text may be: One of the most basic awareness class ser-
struments such as questionnaires. In many cases the dy- vice which can be built upon such a dynamic social network
namic model of these properties of social relations (such model is a visualization which allows to intuitively recog-
nize social distance and group structures. There is a clear
demand for methods and software tools, that are able to ana-
lyze and visualize the evolution of networks [50, p. 1208ff].
Limited by visual and geometric constraints, a few basic
metaphors for temporal or dynamic graphs and networks
have been developed so far, including line graphs with sum-
mary statistics, series or animations of 2D- or 3D-snapshots,
graph overlays, node position tracing and 2,5D or 3D-models
Submitted for review to CHI 2009.
1
with a temporal z axis. But the requirements regarding the based on all this previous work) some approaches on dy-
aspects of visualization in general once formulated by Bran- namic social network visualization (roughly related to our
des [5, p. 7ff] - substance, design and algorithm - can still not approach are e.g. [11, 10, 22, 52, 59, 46] have been pro-
be regarded as solved issues and thus still need to be worked posed. We have compiled a more detailed review of the cited
on. We aim at proposing a solution to dynamic social net- related work in [35].
work analysis that on the hand maintains the mental map
well over time and thus allows a quick overview of the dy-
namics of the relations in a set of people (thus avoiding the PARADIGMS & DESIGN RATIONALS
clutter of some older approaches) as well as an interactive In order to work towards the the goals defined in section ,
UI that allows to focus on more specific aspects and analyze we will combine and adapt several existing techniques in a
them in more depth. unique way and add some new ideas:
Such a method should take the language of social patterns Space-time path We apply Hgerstrand’s “space-time path”
mentioned before and carry it forward into the temporal di- principle [33] to social networks. Euclidean distances in the
mension, and it should combine the “big picture” with ex- two spatial dimensions are derived from social proximity
act metrics of the development of a social network during data which in our Jazz musician network is deduced from
a given period of time, following Ben Shneiderman’s “Vi- co-recordings. Similar approaches are used in physics when
sual Information-Seeking Mantra”: “Overview first, zoom visualizing world-lines of particles in special relativity the-
and filter, then details-on-demand” [55]. Thus such an ap- ory.
proach has to be able to adequately mediate between the so-
cial macro dynamics and micro dynamics. Force-directed layout Social structures in a given time-
slice is visualized by a force-directed layout mechanism, as
The main objective of this paper is to discuss such an in- demonstrated for example by Krempel [41] [42] [43], Dekker
teractive visualization approach and to verify the suitability [15] and others. We use a modified version of the Frucht-
of the technology used and of its visual metaphor by case erman-Reingold algorithm [28], which will be adapted to
studies. support our notion of the “crowd” metaphor following [48].
This original metaphor assumes that most activity within
The application should allow to inspect structures of social a large group takes place in an inner circle which is sur-
networks from the “connectedness perspective”, as defined rounded by the outer fringe with passive people. Important
by Brandes [5, p. 33]. It is intended to support awareness in metrics for this arrangement are the diameter of the circle
communities and in (mobile) social networks with dynami- and the thickness of the ring, permeability and sharpness of
cally changing relations and to give quick visual answers to the borders, and the space between rings, if there are several
questions on varying time- (and thus space-)scales like the ones. This crowd metaphor can be assumed to apply to so-
following: Which are central, important or prominent actors cial micro-contexts (covering small space-time-”intervals”)
and which are peripheral? How does the centrality of actors as well as macro-contexts (covering larger space-time-”in-
develop over time? Are there long-lasting partnerships be- tervals”) and although it has been stated for groups, it can
tween actors? Are there visible structures in the network? be used to intuitively visualize general social structures or
How do structures evolve? How do actor attributes correlate networks.
with visible network structures?
Stacked graphs The inherent temporal graph structure is in-
Proposing convincing approaches as possible solutions to spired by “combined”, “stacked” or “stratified” graph layout
these questions would allow a user to effectively analyze a methods as shown by Erten et al. [23] [24] and Dwyer and
social environment (e.g. personal or professional) and main- Eades [20] [19] and others. Where time is represented by the
tain an overview of the dynamics within such an environ- third dimension of a 3D vizualization.
ment. An interesting field of application are e.g. open inno-
vation processes in dynamically evolving fields (such as life Mental map We use strategies from dynamic graph draw-
sciences) where many actors are involved and such questions ing inspired by solutions described by Branke [8], Diehl et
can be asked with respect to e.g. co-invention-activities. al. [17][16] and Brandes [5] to minimize changes between
subsequent layouts and to preserve the “mental map” [21]
so that the evolution of structures can be followed through
RELATED WORK time. preserving the mental map between time slices is one
There are various approaches to graph drawing (see [47, 30] of the key problems in dynamic graph visualization.
for a good overview) and social network visualization in
general (classic examples are e.g. [36, 5, 15, 57, 43, 38, 31, Tube metaphor We introduce the “tube” metaphor, an en-
42]. Furthermore there are also many approaches to visual- hancement of the “worm” metaphor, which was introduced
ization of dynamic graphs (related to our work are e.g. [12, by Mathews and Roze [45], and enhanced by Dwyer and
13, 7, 17, 16, 8, 27, 6, 20, 23, 24, 14, 18, 1, 9, 25, 19, 29] Eades [20] [19] and Ware et al. [56] to implement the space-
and general time related data (good overview: [2])(related time path. Instead of aggregated cones or simple inter-tem-
to our work are especially [45] and (as a general paradigm) poral edges we use tubular shapes extruded from interpola-
Hgerstrands space-time-paths [33]) (works that are inspired ted NURBS curves to achieve a better compliance with the
by that paradigm: [44, 61, 60, 54, 49])). Furthermore (partly continuity principle of Koffka’s “Gestalt Theory” [40] (cited
2
after [37]). and McFarland [4] [50]. We thus realize a simple way of
mediating between social micro- and macro dynamics. For
a more fine grained mediation the stack of rules for event
aggregation need to be refined further and made adjustable
to the specific social domain.
Time-line and section view A simplified time-line-based
approach is used to show two-dimensional layouts of indi-
(a) Stack of weighted (b) Straight intertempo- vidual “frames” or “time-slices”, similar to the “phase plot”
graphs. ral edges. mechanism by Bender DeMoll and McFarland [3].
Temporal attribute mapping One network-related attribute
can be mapped to the nodes’ spatial coordinates (see above).
Two additional syntactic or semantic actor attributes can be
mapped to the temporal extension of the tubes, one by a
continuous color gradient and the other by radius transition.
Similar approaches have already been suggested [19, p. 101].
(c) Interpolated in- (d) Tubular intertempo-
tertemporal edges. ral edges. Interactive GUI An interactive, explorative three-dimensi-
onal user interface built of a Java3D universe is used, com-
Figure 1. 2.5D Graph stack without and with intertemporal edges. bined with a tabular database view and a two-dimensional
(Time-Dim.:in z-Direction) graph layout. Interactivity is a crucial feature of dynamic so-
cial network visualization since zooming and change of per-
spective are necessary in order to complete one’s overview
Abstraction from nodes and edges The tube metaphor used of the development of a social structure. For a detailed dis-
will abstract completely from graph edges to prevent occlu- cussion of the features of the GUI see [35].
sion, to help focus on the structure and to reveal pure spatio-
temporal movement (spatial proximity corresponds to social
proximity). Omitting the edges for an increased overview is DEFINITIONS, ASSUMPTIONS & REALIZATION
a quite common technique (see e.g. [51]). Our uni-modal dynamic social network model is a temporal
multi-graph G(t) = (V, E(t)) with a set of actors V and
an undirected, weighted time dependent set of edges E(t)
which are known at discrete points in time E(ti ) and are
then interpolated. Each G(ti ) is called a time slice. Each
pair of nodes can be connected by an arbitrary number of
edges. The weights of the edges are normalized to one via
wnorm (e) = w(e)/wmax and will be interpreted as ”social
proximity” values. Furthermore we assume that every node
has a profile which can be modeled as an attribute value pair
list. Such node profiles may contain attributes with slow dy-
(a) Tube-only display (b) Color mapping namics (such as long term interests, fields of study, name,
sex etc.) as well as attributes with fast dynamics (context
parameters) such as location, current activity etc.
From a more abstract point of view the edges also have a pro-
file which contains such elements as the weight of the edge at
a given point in time and the history of instantiation events
of the underlying social relation over time. Thus the edge
weights of a social macro-dynamics perspective are aggre-
gated via domain specific rules from social micro-dynamics
(specific social events or instantiations of social relations re-
spectively).
(c) Color and radius mapping
It is clear that that there are two main perspectives, which
Figure 2. Tube-only model without and with mapping of degree cen- can be applied when analyzing social structures: the view-
trality onto temporal axis. point of connectedness (relational view) and the viewpoint
of profile (entity view) Brandes’ view [5, p. 32ff].
Continuous-time model Temporal attributes are represented The profile viewpoint is realized in our approach by the con-
by a simplified continuous-time model, where social events cept that two real valued attributes (if existing) can be addi-
are aggregated according to rules, similar to (but admittedly tionally visualized in our ”‘tube-only” model via color and
not as flexible as) the model suggested by Bender DeMoll radius of the tubes. The connectedness perspective is at-
3
tributed in our approach through the weight of the edges, Fruchterman-Reingold application for the next slice at ti+1 .
which corresponds to social proximity. There are numerous Other algorithms like Kamada and Kawai [39], would be less
approaches in literature for computing social proximity, for suited for us because of their usage of the graph-theoretical
example either by considering the number of different paths path distance. While Kamada-Kawai is in fact in use for
between two actors [34, ch. 7, p. 9], from the calculation of social network visualization in some applications, the path
geodesic path length[34, ch. 7, p. 14] or as a combination of, length between two nodes is less obvious to interpret that
e.g., weighted adjacency and geodesic distance [15]. the weight of the individual relation. We assume that the ap-
The current version of DySoN assumes that the edge weights plication domain for a dynamic social network visualization
(w(e(ti )) (in the sense of a social macro-context) are be implies a highly directly connected social network where
computed by accumulating social events (in the sense of a two nodes are very likely to be directly connected.
social micro-context) that take place in [ti−1 , ti ] involving In order to further preserve the mental map, we assume that a
the adjacent actors {vl , vm } of e. We assume that these node should remain at it is current position as far as possible
events are situative instantiations of longer lasting social re- if its degree does not change substantially, so we introduce
lations. Naturally, as has been discussed before, the domain an additional attractive force from the nodes position dur-
of application dictates what a good distinction between the ing the FR-run to its position in the previous time slice with
social micro context (social ”events”) and the social macro- a strength proportional to its degree change. Using relative
context (longer lasting social relations) should be. The nest- weight changes instead of degree changes (weights drop to /
ing of social dynamic tiers can be broken down to very fine raise from zero) would be another possibility.
space and time granularities: E.g. from a precise modeling
of single communication acts between two people on a party The original FR algorithm uses a ”spring-paradigm” between
as elementary events of micro-context, continuing with a tier nodes to compute a suitable layout, which uses a repulsive
that aggregated the events from the previous layer and views force fr = −k 2 /δ and an attractive force fa = δ 2 /k be-
the event of visiting the party alongside the events of visit- tween two nodes, where δ is the euclidean distance between
ing the zoo together two days later as elementary events and them and k (being roughly analogous to the spring constant
finally a layer that aggregates these events over a month and or ”natural length of the spring” is a simple function of the
views months of intense interaction as elementary events. visualization canvas dimensions w and h and some experi-
Each tier (coarser time / space granularity) aggregates re- mental constant c). The forces are directed along the vec-
lation instantiations (events) from a lower tier (finer time / tor from node one to node two. We modify the original
space granularity). approach by several means. First we introduce our edge
The investigation of this type of modeling is subject to cur- weights by proportionally strengthening the attractive force
rent research of one of the authors. We aim at investigating fa0 = fa ∗ w(e). The second modification introduces an
the modeling assumption that the intensity and other aspects additional attractive gravitational force (inspired by [26]) to
of higher tier relations can be readily deduced from the prop- the center of the slice canvas. This accounts for the effect
erties of lower tier relation events. occurring with pure FR, that isolated nodes are pushed to
the canvas borders by the lack of attractive force. In order
It is possible to substitute social proximity with (profile- to emphasize the impact of centrality, our additional attrac-
)similarity depending on the application domain. Analo- tive gravitational force is fg = δ 2 ∗ (deg(v) + d)/k where
gous to the nesting of social proximity with respect to time deg(v) is the degree of v, δ its distance to the center and d
and space granularity (social micro- / macro-context) one an additional steering parameter.
can use for the similarity calculation profile parameters with
high dynamics (such as current micro-location) or averaged The complexity of the original FR algorithm has been stated
variants of these (e.g. coarse area of usual ”residence”). as Θ(|V |2 + |E|) [28, p. 1138] and our complete layout al-
Again as in the case of relational dynamics we assume that gorithm can be shown [35] to have an overall complexity of
the macro-tier contains averages of the next lower micro- O(|V |2 s) where s is the number of time slices. So our mod-
tier. Some Profile elements with very slow dynamics (such ifications to not add substantially to the overall time com-
as name) that are not averaged versions of related profile el- plexity. It may be worthwile to study the effects of intro-
ements with higher dynamics are often less useful for simi- ducing node ”inertia” for mental map preservance and using
larity calculations. the result layout from the previous time slice as the starting
layout for the modified FR algorithm in the current layout
One of the main goals for the relative layout of the planar with respect to the number of iterations it takes the modi-
graphs corresponding to each time slice is that the layout of fied FR-Algorithm to converge in each time slice. One can
consecutive time slices is supposed to preserve the ”‘men- expect a substantially decreased number of iterations, since
tal map” [21] as much as possible. This paradigm inter- we start from a ”good previous solution” and limit the ”node
feres with mapping the social distances as exactly as pos- mobility”.
sible. We solve these conflicting demands by using a modi-
fied Fruchterman-Rheingold [28] (FR) layout algorithm for Having computed the positions of each node in each time
each slice as a compromise, also because it is easy to adapt slice, these points have to be interpolated with a suitable
to our purposes (edge weights as measures of social prox- smooth curve (∈ C 2 (see [58] for an easy motivation)) which
imity). We furthermore use the standard approach of us- is the center of the tube for that particular node (actor). We
ing the resulting layout of slice ti as initial layout for the evaluated interpolation polynomials, Bezier curves and sim-
4
Figure 3. Decay of the Miles Davis Band in the early seventies.
ple B-splines for the purpose and found severe drawbacks for contributions for a given time slice is the domain specific
each [35] and arrived at NURBS (Non Uniform Rational B- heuristic rule, which connects social micro-dynamics (bro-
Splines) [53] of degree 3 as the best choice for our problem. ken down to individual instantiations of social ”co-recording”
See [35] on how we compute knot points and control points relations) and macro-dynamics (e.g. when applying a time
or these curves. We then build our tube surfaces as cylindri- slice of one year. Depending on the domain of interest these
cal NURBS surfaces around the interpolation curves. rules need to be adapted.
Concerning the ”profile” dimensions color and radius we We made substantial efforts to avoid counting re-releases.
chose (for the current prototype) to visualize node degree The color corresponds to the node degree as explained be-
with color and radius, because n our paradigm the edges are fore and the tube radius is also set to reflect the node degree
missing completely. The color paradigm is to chose ”hot to support the color coding.
colors for (nodes) tubes with a high node degree (these are
perceived to be ”socially active” in the given time slice) and Figure 3, for example, depicts the breakup of a band which
also to give them larger radii the more connected they are. played together for some years. The involved musicians all
[35] describes the details of these calculations. started solo careers and their own band projects after one
successful key recording with the band leader. You see the
effect that tubes are crossing here, though the clique has not
STUDY: THE JAZZ-NETWORK
changed, which has to be addressed by improving the incre-
As a first step to verify the suitability of the approach we col-
mental layout algorithm.
lected an extensive dataset on musical collaborations in Jazz
Figure 4 shows an actor, who has been central for some years
and checked from our own pre-existing knowledge of the
before relations to the other participating actors break. This
Jazz-scene whether the tool was able to fulfill the goals. We
happened due to a couple of solo recordings, which do not
crawled on of the numerous publicly available, Wiki-style
provide the actor with high centrality.
(socially crafted) discography data-base Discogs (www.dis-
Our findings with several other examples were, that the sys-
cogs.com) with a snowball approach [34] and substituted
tem was able to meaningfully visualize phenomena in the
missing biographical data of the musicians by a supplemen-
Jazz scene over the last decades. A further evaluation would
tal crawling process of Wikipedia. This resulted in 96798
have to empirically manifest this claim by doing an exten-
musicians who played on 224173 tracks on 37773 albums.
sive study with a set of Jazz experts.
Thus the resulting social network is in fact a two mode net-
Future research will also investigate the assumption that the
work (mode one: musicians, mode two: albums). Each mu-
system’s basic design metaphors are also suitable for the
sical co-contribution of two musicians for a track is viewed
vizualization of social relations on a shorter time scale (e.g.
as an event and accumulated to the temporal weight of the
as part of awareness services in mobile social network appli-
respective edge in the respective time slice. Adding up these
5
Figure 4. Actor with period of high degree centrality.
cations). glect a fine grained dynamic social network model and, in
general, the user has to take care for himself about modeling
UI/PARAMETERS OF THE CURRENT DYSON APPLICA- his social relations. Truly mobile social network models and
services which are socially sensitive as well as context sensi-
TION tive will allow for semi-automatic detection and modeling of
Besides several mainteneance UI elements (e.g. for manip- social relations on all time / space scales (the social micro-
ulation the actor set), DySoN allows to interactively control and macro-contexts) and will make intimate use of the re-
and manipulate the visualization in order to analyze certain sulting ”multi-dynamic” social network. A possible service
details. The user can turn and zoom the visualization in sev- could be e.g. a ”socio-context-cast” communication service
eral ways. Furthermore the geometry of the tubes can be that allows to publish messages to certain combined social
manipulated by e.g. setting the overall tube diameter (e.g. and physical contexts (”Send a message to all persons that
depending on the number of actors). Lighting and color attended Jim’s party yesterday”).
models can also be adjusted: E.g. Degree Transparency is Because of the highly dynamic nature of social micro-contexts,
one approach to improve the readability of the model (es- it is not reasonable to assume that continuously uploading
pecially those with high numbers of actors and thus a high such information to a central social networking platform is
density of tubes) by making those parts of the tubes trans- reasonable. Furthermore the more fine grained the collected
parent, that represent a time period with a degree of less than social information is, the more pressing are the privacy is-
the value set by the Threshold slider. The amount of trans- sues connected with an unconstrained publication of such
parency can be adjusted by the Opacity slider. Furthermore data. Instead of such an Orwellian central platform, one
the user can interactively determine the profile parameters would switch to a decentralized social networking paradigm
that are mapped to the radius and color temperature. such as NoseRub.
There a lot of other parameters which can be adjusted. See Clearly, awareness services are one of the key features of
[35] for details. such a mobile P2P style social networking framework. DySoN
can be used to visualize either social proximity (”what’s the
APPLICATION OF DYSON AS AWARENESS SERVICE EL- (micro-/meso-/macro-dynamics of my ‘friendship’ relations?”)
EMENT IN MOBILE P2P SOCIAL NETWORKS or social similarity (e.g. co-locatedness (”who is in my vicin-
Existing social network platforms such as Facebook often ity?”)) on all time scales.
are mainly focused on long term social relations, thus at the In order to do that, we have to work on heuristics that ag-
social macro-context. What is happening on a finer time gregate the events on the micro-scale into social relationship
scale (except for some aspects such as Online-awareness etc) strengths (or proximity values) on the next higher level.
is only covered by some platforms such as Twitter. But these
platforms often tend to focus on the individual user and ne-
6
SUMMARY, DISCUSSION AND FUTURE WORK Proceedings of the 10th ACM Conference on Hypertext
We discussed a novel method to visualize dynamic social and hypermedia: returning to our diverse roots, pages
networks. A case study of collaborating Jazz musicians re- 51–60, 1999.
vealed that the approach indeed matches the goals that were
formulated in section . On a more general level an empiri- 11. C. Chen and L. Carr. Visualizing the evolution of a
cal user study would have to be conducted. Since it is very subject domain: a case study. In VIS ’99: Proceedings
hard to measure the ”quality” of a visualization the design of the conference on Visualization ’99, pages 449–452,
of such a study would have to involve standardized data and Los Alamitos, CA, USA, 1999. IEEE Computer
a comparison with other approaches which is difficult since Society Press.
every existing other approach aims at slightly different as- 12. E. H. Chi. Web analysis visualization spreadsheet.
pects. Since the dynamics of social networks is coming more Technical report, Xerox Palo Alto Research Center,
more to the focus of attention (especially due to mobile in- Palo Alto, CA 94304, 1999.
teraction paradigms), the problem of useful dynamic social
network visualization still remains an interesting topic for 13. E. H. Chi and S. K. Card. Sensemaking of evolving
the future. web sites using visualization spreadsheets. In INFOVIS
’99: Proceedings of the 1999 IEEE Symposium on
REFERENCES Information Visualization, page 18, Washington, DC,
1. A. Ahmed, T. Dwyer, C. Murray, L. Song, and Y. X. USA, 1999. IEEE Computer Society.
Wu. Wilmascope graph visualisation. In INFOVIS ’04: 14. C. Collberg, S. Kobourov, J. Nagra, J. Pitts, and
Proceedings of the IEEE Symposium on Information K. Wampler. A system for graph-based visualization of
Visualization (INFOVIS’04), page 216.4, Washington, the evolution of software. In SoftVis ’03: Proceedings
DC, USA, 2004. IEEE Computer Society. of the 2003 ACM symposium on Software visualization,
2. W. Aigner, S. Miksch, W. Müller, H. Schumann, and pages 77–ff, New York, NY, USA, 2003. ACM Press.
C. Tominski. Visualizing time-oriented data – a 15. A. Dekker. Conceptual distance in social network
systematic view. Computers + Graphics, analysis. Journal of Social Structure, 6(3), 2005.
31(3):401–409, June 2007.
16. S. Diehl and C. Görg. Graphs, they are changing –
3. S. Bender-deMoll and D. A. McFarland. Interaction, dynamic graph drawing for a sequence of graphs. In
time, and motion: Animating social networks with Graph Drawing: 10th International Symposium, GD
sonia, 2004. 2002, Irvine, CA, USA, August 26-28, 2002. Revised
Papers, volume 2528/2002 of Lecture Notes in
4. S. Bender-deMoll and D. A. McFarland. The art and Computer Science, pages 23–30. SpringerLink, 2002.
science of dynamic network visualization. Journal of
Social Structure, 7(2), 2006. 17. S. Diehl, C. Görg, and A. Kerren. Preserving the mental
map using foresighted layout. In Proceedings of the
5. U. Brandes. Layout of Graph Visualizations. PhD Joint Eurographics-IEEE TVCG Symposium on
thesis, Universität Konstanz, 1999. Visualization, VisSym, pages 175–184. Springer Verlag,
6. U. Brandes and S. R. Corman. Visual unrolling of Wien, New York, 2001.
network evolution and the analysis of dynamic 18. T. Dwyer. A scalable method for visualising changes in
discourse. In INFOVIS ’02: Proceedings of the IEEE portfolio data. In APVis ’03: Proceedings of the
Symposium on Information Visualization (InfoVis’02), Asia-Pacific symposium on Information visualisation,
page 145, Washington, DC, USA, 2002. IEEE pages 17–25, Darlinghurst, Australia, Australia, 2003.
Computer Society. Australian Computer Society, Inc.
7. U. Brandes and D. Wagner. A bayesian paradigm for 19. T. Dwyer. Two and a Half Dimensional Visualisation of
dynamic graph layout. In Proceedings of the 5th Relational Networks. PhD thesis, University of Sydney,
International Symposium on Graph Drawing (GD’97), 2005.
volume 1353 of Lecture Notes in Computer Science,
pages 236–247. Springer Verlag, 1997. 20. T. Dwyer and P. Eades. Visualising a fund manager
flow graph with columns and worms. In Proceedings of
8. J. Branke. Dynamic graph drawing. Lecture Notes In the 6th International Conference on Information
Computer Science, pages 228–246, 2001. Visualisation, IV02, pages 147–158. IEEE Computer
Society, 2002.
9. C. Chen. The centrality of pivotal points in the
evolution of scientific networks. In IUI ’05: 21. P. Eades, W. Lai, K. Misue, and K. Sugiyama.
Proceedings of the 10th international conference on Preserving the mental map of a diagram. In
Intelligent user interfaces, pages 98–105, New York, Proceedings of Compugraphics, pages 34–43, 1991.
NY, USA, 2005. ACM Press.
22. T. Erickson and M. R. Laff. The design of the ’babble’
10. C. Chen and L. Carr. Trailblazing the literature of timeline: a social proxy for visualizing group activity
hypertext: author co-citation analysis (1989–1998). In over time. In CHI ’01: CHI ’01 extended abstracts on
7
Human factors in computing systems, pages 329–330, 36. J. Heer. Exploring Enron – Visualizing ANLP Results,
New York, NY, USA, 2001. ACM Press. 2004.
23. C. Erten, P. J. Harding, S. G. Kobourov, K. Wampler, 37. K. Hinum. Gravi++ - An Interactive Information
and G. Yee. Exploring the computing literature using Visualization for High Dimensional, Temporal Data.
temporal graph visualization, 2003. PhD thesis, Vienna University of Technology, Institute
of Software Technology and Interactive Systems,
24. C. Erten, P. J. Harding, S. G. Kobourov, K. Wampler, Favoritenstrasse 9-11/186, A-1040 Vienna, Austria,
and G. V. Yee. Graphael: Graph animations with December 2006.
evolving layouts. In Graph Drawing, volume 2912 of
Lecture Notes in Computer Science, pages 98–110. 38. M. Höpner and L. Krempel. The politics of the german
Springer Verlag, 2004. company network, September 2003.
25. C. Erten, S. G. Kobourov, V. Le, and A. Navabi. 39. T. Kamada and S. Kawai. Automatic display of
Simultaneous graph drawing: Layout algorithms and network structures for human understanding. Technical
visualization schemes. Journal of Graph Algorithms Report 88-007, Department of Information Science,
and Applications, 9(1):165–182, July 2005. University of Tokyo, 1988.
26. A. Frick, A. Ludwig, and H. Mehldau. A fast adaptive 40. K. Koffka. Principles of Gestalt Psychology. Harcourt,
layout algorithm for undirected graphs. In R. Tamassia Brace, & World, 1935.
and I. G. Tollis, editors, GD ’94: Proceedings of the 41. L. Krempel. Simple representations of complex
DIMACS International Workshop on Graph Drawing, networks: strategies for visualizing network structure.
volume 894, pages 388–403, London, UK, 10–12 1994. Presented at the 3rd European Conference for Network
Springer-Verlag. Analysis, Munich, 1993.
27. C. Friedrich and P. Eades. Graph drawing in motion. 42. L. Krempel. Visualizing networks with spring
Journal of Graph Algorithms and Applications, embedders: Two-mode and valued graphs. Presented at
6(3):353–370, February 2002. the Joint Statistical Meetings of the American
Statistical Association, August 1999.
28. T. M. J. Fruchterman and E. M. Reingold. Graph
drawing by force-directed placement. Software - 43. L. Krempel. The language of networks. Presented at the
Practice and Experience, 21(11):1129–1164, 1991. Ars Electronica 2004, Linz, September 2004.
29. M. Gaertler and D. Wagner. A hybrid model for 44. M.-P. Kwan and J. Lee. Geovisualization of human
drawing dynamic and evolving graphs. In S. Leonardi, activity patterns using 3d gis: A time-geographic
F. M. auf der Heide, and D. Wagner, editors, approach. In M. F. Goodchild and D. G. Janelle,
Algorithmic Aspects of Large and Complex Networks, editors, Spatially Integrated Social Science: Examples
number 05361 in Dagstuhl Seminar Proceedings. in Best Practice, chapter 3. Oxford University Press,
Internationales Begegnungs- und Forschungszentrum Oxford, 2003.
fuer Informatik (IBFI), Schloss Dagstuhl, Germany,
2006. 45. G. Matthews and M. Roze. Worm plots. IEEE
Computer Graphics and Applications, 17(6):17–20,
30. R. T. I. G. T. Giuseppe Di Battista, Peter Eades. Graph November-December 1997.
Drawing - Algorithms for the Visualization of Graphs.
46. D. A. McFarland and S. Bender-deMoll. SoNIA (Social
Prentice Hall, 1999.
Network Image Animator), 2007.
31. P. Gleiser and L. Danon. Community structure in jazz. 47. D. W. E. Michael Kaufmann. Drawing Graphs -
Technical report, Universitat de Barcelona, 2003. Methods and Models. Springer, 2001.
32. G. Groh. Ad-Hoc-Groups in Mobile Communities - 48. S. Milgram. The individual in a social world - Essays
Detection, Modeling and Applications. PhD thesis, and Experiments. Series in social psychology.
2005. Addison-Wesley, Reading, Massachusetts, 1977.
33. T. Hägerstrand. What about people in regional science? 49. H. J. Miller. Activities in space and time. In K. H.
Papers of the Regional Science Association, 24:7–21, P. Stopher, K. Button and D. Hensher, editors,
1970. Handbook of Transport 5: Transport Geography and
Spatial Systems. Elsevier Science, 2004.
34. R. A. Hanneman and M. Riddle. Introduction to social
network methods, 2005. 50. J. Moody, D. McFarland, and S. Bender-deMoll.
Dynamic network visualization. American Journal of
35. H. Hanstein. Interactive visualization of dynamic social Sociology, 110(4):1206–1241, January 2005.
networks. Master’s thesis, Fakultt fr Informatik, Prof
Schlichter, Technische Universitt Mnchen, Munich, 51. P. Mutton. Inferring and visualizing social networks on
Germany, 2007. internet relay chat, 2004.
8
52. P. Mutton. Piespy social network bot – inferring and
visualizing social networks on irc, 2006.
53. L. A. Piegl. On NURBS: A Survey. IEEE Computer
Graphics and Applications, 11(1):55 – 71, January
1991.
54. S.-L. Shaw. Time geography for activity modeling with
gps tracking data. Presentation at FHWA-sponsored
Peer change and CSISS Specialist Meeting, Santa
Barbara CA, October 10–11 2005.
55. B. Shneiderman. The eyes have it: A task by data type
taxonomy for information visualizations. In IEEE
Visual Languages, pages 336–343, College Park,
Maryland 20742, U.S.A., 1996.
56. C. Ware. GeoZui4D – GEOreferenced Zooming User
Interface 4D, 2007.
57. R. Wengenmayr. Thinking with your eyes.
MaxPlanckResearch, 2:64–71, 2005.
58. Wikipedia. Nonuniform rational B-spline, July 2007.
59. R. Xiong and J. Donath. Peoplegarden: creating data
portraits for users. In UIST ’99: Proceedings of the
12th annual ACM symposium on User interface
software and technology, pages 37–44, New York, NY,
USA, 1999. ACM Press.
60. H. Yu. Spatio-temporal gis design for exploring
interactions of human activities. Cartography and
Geographic Information Science, 33(1):3–19, 2006.
61. H. Yu and S.-L. Shaw. Revisiting hägerstrand’s
time-geographic framework for individual activities in
the age of instant access. In Proceedings of Research
Symposium on Societies and Cities in the Age of Instant
Access, Salt Lake City, Utah, November 10–12 2005.
9