Volume xx (200y), Number z, pp.

1–28

Visual Analysis of Large Graphs:

State-of-the-Art and Future Research Challenges

T. von Landesberger1,2 , A. Kuijper1,2,3 , T. Schreck1 , J. Kohlhammer2 , J.J. van Wijk4 , J.-D. Fekete5 , and D.W. Fellner1,2,3

1 Technische Universität Darmstadt, Germany

2 Fraunhofer IGD, Darmstadt, Germany
3 Technische Universität Graz, Austria
4 Technische Universiteit Eindhoven, the Netherlands
5 INRIA, France

Abstract

The analysis of large graphs plays a prominent role in various fields of research and is relevant in many important
application areas. Effective visual analysis of graphs requires appropriate visual presentations in combination
with respective user interaction facilities and algorithmic graph analysis methods. How to design appropriate
graph analysis systems depends on many factors, including the type of graph describing the data, the analytical
task at hand, and the applicability of graph analysis methods. The most recent surveys of graph visualization and
navigation techniques cover techniques that had been introduced until 2000 or concentrate only on graph layouts
published until 2002. Recently, new techniques have been developed covering a broader range of graph types,
such as time-varying graphs. Also, in accordance with ever growing amounts of graph-structured data becoming
available, the inclusion of algorithmic graph analysis and interaction techniques becomes increasingly important.
In this State-of-the-Art Report, we survey available techniques for the visual analysis of large graphs. Our re-
view firstly considers graph visualization techniques according to the type of graphs supported. The visualization
techniques form the basis for the presentation of interaction approaches suitable for visual graph exploration. As
an important component of visual graph analysis, we discuss various graph algorithmic aspects useful for the
different stages of the visual graph analysis process. We also present main open research challenges in this field.
Categories and Subject Descriptors (according to ACM CCS): Data Structures [E.1]: Graphs and Networks;
Trees—Mathematics of Computing [G.2.2]: Discrete Mathematics—Graph Theory Information Systems [H.4]:
Applications— Information Systems [H.5.2]: Interfaces and Presentation—User Interfaces

1. Introduction concentrates on the development of effective graph layouts

and visual mappings. The visualization of large graphs is ac-
The analysis of graphs is important in many application ar- companied by effective interaction techniques, in particular,
eas including finance, biology, sociology, transportation, and in cases when the whole graph is too complex or large to be
software engineering; It includes a variety of different tasks. visualized in one static view. The interaction alone may not
The main aspects relate to the understanding of global and be sufficient to accomplish certain analytical tasks. There-
local structure of the graph, the connections between en- fore, algorithmic support — such as machine learning, or
tities, the clusters of highly connected entities, etc. Such graph analysis algorithms — needs to be supported in inter-
high level tasks often consist of a series of low level tasks active visualization systems. Such integrated visual analy-
[LPS∗ 06], in particular when dealing with large and com- sis of large data sets is the main focus of the research field
plex graphs. called Visual Analytics, which evolved from Information Vi-
The analysis of graphs is often supported by their visual sualization and Scientific Visualization [KMS∗ 08]. It has ef-
presentations. In this respect, graph visualization research fectively started to grow after the publication of the seminal

submitted to COMPUTER GRAPHICS Forum (1/2011).

2 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

visual analysis of graphs applicable across various domains.

Visual graph Owing to the broad scope of the article, we present the main
representation features (strengths and weaknesses) of the techniques as far
(Sec. 3)
as they were discussed by the authors of the papers or were
mentioned in evaluations. In particular cases, we point out
(Sec. 5)
the readers to evaluation papers for more details.
Graph User
algorithmic interaction
analysis (Sec. 4) User interaction
(Sec. 2) Visualization
3(Sec. 4)
Mapping
2
(Sec. 3) Transformation
Figure 1: The main components of visual graph analysis 1 (
(Sec. 2)
)

considered in this report. Data Knowledge

Model Model
b ildi
building visualization
i li ti
4 (Sec. 5)

Models
book by Thomas and Cook in 2005 [TC05]. Therein, Visual Parameter
Analytics is defined as the science of analytical reasoning refinement

facilitated by interactive visual interfaces. Recently, Visual Information mining

Analytics has been a major driving force for the research and Feedback loop
development of interactive visualization techniques for large
amounts of data including graphs. Figure 2: The Visual Analytics process by Keim et al.
[KAF∗ 08] with the four key steps 1: data pre-processing,
Our motivation for this report is two-fold. First, we rec- 2: mapping/layout 3: visual user interaction 4: model-based
ognize that by now most recent graph visualization sur- c
analysis. 2008 Springer-Verlag Berlin Heidelberg.
vey [HMM00, DPS02] date back several years. Therefore,
we aim to provide an update by adding more recent publi-
cations to the body of work presented in these surveys. Sec- The structure of the report reflects the steps of the visual
ond, we aim to take a Visual Analytics perspective on the analytics process introduced by Keim et al. [KAF∗ 08] (see
field of visual graph analysis by explicitly considering in a Figure 2). Section 2 details on definitions and a classifica-
unified way the aspects of visual representation, user inter- tion of graphs by types and introduces main pre-processing
action, and algorithmic analysis (see Figure 1). These three methods for visual graph analysis. This section is the basis
elements form the basis for effective visual graph analysis for a discussion of visual graph representations given in Sec-
systems and are closely interrelated. For example, algorith- tion 3. Sections 4 and 5 survey key approaches for interac-
mic graph analysis may be applied as a preprocessing step tion with and algorithmic analysis of graphs in visualization,
before a specific graph layout is determined for visual repre- respectively. Finally, Section 6 concludes and outlines future
sentation. Interactive direct object manipulation approaches challenges in this research domain.
are often useful for exploring large and complex graphs vi-
sually. Also, by means of user interaction, further graph an-
alytic processing steps, or updates to the presented views, 2. Basic Graph Definition and Preprocessing
can be requested. The algorithmic analysis thereby helps to Techniques
reveal interesting aspects of the data. The user involvement In this section, we recall fundamental graph definitions as
in this analytic process can vary from an automatic analy- well as approaches for graph preprocessing useful for subse-
sis, where the calculation is done without user involvement, quent graph visualization.
over a user-driven analysis, where the user triggers the algo-
rithmic processing of the data, up to a user-steered process
where the user has full control over the analytical process 2.1. Definitions
including setting the algorithm parameters.
Graphs are a prominent data structure within Visual Analyt-
In this state-of-the-art report, we provide a systematic ics and related research fields. Often, graphs are applied for
overview of the main approaches in each of the three aspects describing relationships between entities. A graph refers to
of visual analysis of graphs. We therefore develop a classifi- a set of vertices (nodes) and a set of edges (i.e., links) that
cation of techniques according to these aspects. Within each connect pairs of vertices. It is a pair G = (V, E); E ⊆ [V 2 ],
category, we refer to exemplary papers, while focusing on where elements of V are vertices and elements of E are
new developments in the visual graph analysis area. We dis- edges [Die05]. Furthermore, attributes can be attached to
cuss the respective techniques, in order to offer the reader vertices and edges, e.g., to denote their type, size, or some
the possibility to concentrate on methodological aspects of other application related information.

submitted to COMPUTER GRAPHICS Forum (1/2011).

 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 3

Graphs are often classified into undirected and directed

[HMM00]. For a directed graph (resp. undirected), the edge Time
Structure
vertices e = (v1 , v2 ) are ordered (resp. unordered). A graph dependence
containing both directed and undirected edges is called
mixed. Static Trees Hierarchy
g p
graphs
A path of length s in G is a sequence of connected vertices Directed
(Generic)
pathG (a1 , as ) = a1 , a2 , . . . , as where ai ∈ V and (ai , ai+1 ) ∈ Time-
Attribute change
Graphs
Undirected
Mixed
E. A cycle is a closed path with a1 = as . A tree is a con- dependent Structure change

nected undirected graph without cycles [Die05]. A Tree T graphs Attribute &
structure change Compound
is called rooted when one vertex r is distinguished as a so graphs
called root node: T = (V, E, r). Such trees are often treated as
hierarchies, where the length of the path to the root denotes Figure 3: Classification of graphs according to their time
the level of a vertex in the hierarchy. However, formally, a hi- dependence and graph structure.
erarchy is a directed acyclic graph so, in a formal hierarchy,
a node can have several paths to the root node. In this survey,
we use the the term hierarchy as synonym to “rooted tree”.
Note that a connected graph can be transformed to a tree
Furthermore, graphs may be distinguished according to
by removing edges causing cycles while keeping the graph
their topological properties. There exists a variety of litera-
connected.
ture on graph theory (e.g., [Die05]) which focuses on graph
In graph theory literature, a directed graph with weighted terminology, classification, and algorithmic graph analysis.
edges is also called a network. In information visualization, In the following, we mention only the most relevant termi-
the term network is often used in a broader sense denoting a nology used later in this report. Basic graph properties in-
graph with attributes associated with vertices and edges. clude the number of nodes, graph density, and connectivity.
An additional graph category are so-called compound Properties are often taken into account (or are a prerequisite)
graphs. A compound graph C = (G, T ) is defined as a graph for certain visualization techniques. These properties often
G = (V, EG ) and a rooted tree T = (V, ET , r) that share the heavily influence which visualization methods can be used
same set of vertices, such as: or fall short, with respect to readability and performance.
For example, the increasing number of nodes, higher graph
∀e = (v1 , v2 ) ∈ EG , v1 6∈ pathT (r, v2 )andv2 6∈ pathT (r, v1 ) density, or both pose a scalability problem in visualization
Relationships between vertices are expressed by T : vertices owing to limited display space and human perception capa-
sharing a common parent in T belong to the same “group”. bilities.
When two vertices sharing a common parent are connected The number of nodes (i.e., graph order) is often referred to
in G, they share a generic relationship. Many other kinds as graph size (|V |). Graph density is the number of edges
of relationships can be expressed including hierarchic and relative to the maximum potential number of edges D =
2|E|
cross-group. |V |(|V |−1)
. Sparse graphs have around O(|V |) < |E| <<
Compound graphs can be created by successive aggre- O(|V |2 ) edges, while dense graphs show density values close
gation (or clustering) of graph vertices in a bottom-up ap- to one. Graphs with the maximum number of edges are
proach. This operation usually involves creating new nodes called complete graphs. A clique is a subset of a graph that
as group/cluster parents. In this case, vertices (and implicitly, is fully connected.
also edges) of the original graph are aggregated (i.e., added According to the graph size, graphs are often referred to,
as children of the group parent), thereby creating constructed e.g., as small or large. The definition of large graphs is how-
meta-nodes or super-nodes. The attributes of the meta-nodes ever not standardized. Often graphs with thousands, hun-
can be calculated from the attributes of the merged nodes. dreds of thousands or millions of nodes are called large.
Similarly, edges between meta-nodes are aggregated into However, not only the number of nodes determines the no-
meta-edges and their attributes can be calculated from the tion of a “large” graph. Graph density and connectivity also
original edges. Compound graphs which are constructed in play an important role for the notion of a “large” graph. From
this way are also referred to as aggregated graphs. The list the visualization point of view, “large” graphs usually lead
of operations that can be performed on such graphs is depen- to cluttered displays. In algorithmic analysis, “large” graphs
dent on the particular application and graph type. refer to long computational times or memory footprint larger
than the available RAM size. A discussion about the influ-
Graphs may also evolve over time, thereby forming dy-
ence of graph size and density on visualization and construc-
namic graphs (i.e., time-dependent graphs) in contrast to
tion of graphs for testing visualizations according to these
static graphs. Time-dependent changes may affect the at-
parameters is provided in [Mel06].
tributes of nodes and edges, the graph structure, or both. Fig-
ure 3 summarizes the graph classification presented above. Several special graph structures appear often in real-

submitted to COMPUTER GRAPHICS Forum (1/2011).

4 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

world cases, and dedicated visualization methods have been

developed for these [ACJM03,vHW08,JHGH08,MJW∗ 09].
For example, social networks usually exhibit a structure
called small world network: the typical distance between two
nodes grows proportionally to log |V |. Scale-free networks,
e.g., protein networks or certain types of social networks
have a degree distribution that follow approximately the
power law. Bipartite graphs are graphs whose nodes form
two disjoint sets: V1 ,V2 with V1 ∪ V2 = V and V1 ∩ V2 = ∅,
such that: ∀e = (v1 , v2 ) ∈ E, v1 ∈ V1 andv2 ∈ V2 . (a) Original graph (b) Stochastic edge sampling

2.2. Algorithmic Graph Preprocessing

In graph visualization, algorithmic graph preprocessing of-

ten includes graph simplification to reduce the size, while
maintaining the main graph structure. Also pre-processing of
graph properties can be used for graph visualization (in algo-
rithms for positioning of nodes and edges) or highlighting of
interesting parts of the graph. The modified graph is used
then for an easier visual inspection as large and complex (c) Geodesic clustering (d) Structure-based filtering
graphs are difficult to understand even using advanced node
and edge positioning algorithms (layouts). Such preprocess- Figure 4: Example of various graph reduction techniques.
ing steps can usually be performed automatically without The graphs are visualized using the GEM layout [JHGH08].
user interaction. There are two main approaches to graph re- c
2008 IEEE.
duction: graph filtering and graph aggregation.

Graph filtering There are two types of filtering: stochas-

tic and deterministic. Stochastic filtering is mainly based
on random selection of nodes and edges from the orig-
inal graph. These methods are compared in [LF06]. De-
terministic filtering uses, as its name suggests, a deter-
ministic algorithm for the selection of the nodes/edges to
be removed. This filtering can be based on node/edge at-
tributes, on topologic values such as betweenness central-
ity, or other graph properties. For example, filtering based
on edge-betweenness-centrality can be used for removal of
less important edges while keeping the underlying structure
(connectedness and other features such as cliques) of the
graph [JHGH08] (see Figure 4).

Graph aggregation In this approach, nodes and edges are Figure 5: Graph aggregation for multi-scale graph visual-
merged to single nodes and edges, thereby reducing the size ization [EDG∗ 08]. 2008
c IEEE. Top: Graph aggregation
of the graph and revealing relationships between groups of schema showing several levels of aggregation. Darker rect-
nodes. Graph aggregation can be repeated multiple times, angles show the corresponding data areas in the aggrega-
creating a hierarchical graph, which is a special kind of tion. Bottom: Example of graph aggregation using a matrix
compound graph. There are various ways of aggregating a visualization.
graph, including using predefined node hierarchies, or ag-
gregation according to node attributes, or according to the
node clusters [EDG∗ 08, BDL∗ 10], to name a few. Figure 5
3. Visual Representations of Graphs
(top) shows an example aggregation schema with several ag-
gregation levels. The highlighted rectangle shows the corre- Visualization is one of the main means of exploratory graph
sponding data in each aggregation level. Figure 5 (bottom) analysis. It includes the development of appropriate types of
shows the original and aggregated data in a matrix visualiza- visual representations (e.g., matrix or node-link diagrams),
tion. efficient placement of graph elements on the screen and ef-

submitted to COMPUTER GRAPHICS Forum (1/2011).

 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 5

ficient visual attribute mappings (design of graph elements Figure 6). There have been several studies comparing the dif-
for improved readability of the drawing). ferent ways of visualizing trees [SCGM00, BN01, vHvW02,
Kob04, AK07]. A very useful visual overview of tree visual-
In computer-created graph visualization, several so-called
ization has been provided in the poster [JS10]. It is difficult
aesthetic criteria are taken into consideration. They are
to unify these results as they differ significantly. Recently,
usually implemented as objective functions to optimize in
Ziemkiewicz and Kosara have shown that the effectiveness
layout algorithms. The standard criteria include minimizing
of the visualization technique depends not only on the task to
the number of crossings, minimizing the total drawing area,
be solved, but also on the formulation of the task assignment,
maximizing symmetries and many more related to particular
i.e., if it reflects a containment or a level metaphor [ZK08].
types of graphs and edge drawing styles [Pur97, DBETT99,
BBD09]. Recently, Beck et al. [BRSG07] extended previ-
ous works to focus on both static and dynamic graphs irre-
spective of their graphic representations (including also ma-
trix representations in addition to node-link diagrams). They
consider three groups of criteria: general, dynamic and aes-
thetic scalability.
• The general criteria include reduction of visual clutter,
reduction of spatial misunderstanding resulting from spa-
tial closeness, maximization of spatial matching of items (a) Node-link diagram (b) Space-filling diagram
for following paths and maximization of space efficiency.
• For dynamic graphs, the following criteria are desired:
maximization of display stability between time points, re-
duction of cognitive load when analyzing time dynamics,
minimization of temporal aliases mainly owing to posi-
tioning of different nodes in the same place in two time
periods.
• Aesthetic scalability criteria refer to graph readability for
larger graphs, i.e., scalability in number of vertices (i.e,
increasing graph order), scalability in number of edges (c) Combined representation
(i.e., increasing graph density), and scalability in number
of graphs, in particular with increasing number of time Figure 6: Three types of hierarchy visualization techniques.
steps for which graph data is given. c
a) Node-Link, b) space-filling, c) hybrid. [ZMC05], 2005
IEEE.
All these criteria are important but they cannot be simul-
taneously optimized and are not sufficient to design a good
layout which is usually data and task dependent. Therefore,
Node-link techniques These approaches use links between
exploratory graph visualization requires more than one lay-
items to depict their relationship. Layout algorithms con-
out algorithm to reveal the several perspectives on relation-
trolled by optimization criteria or the node positions. Many
ships between nodes.
layout algorithms have been proposed to date in the Graph
In this section, we describe the main graph visualization Drawing community. They include layered, radial or balloon
techniques following the graph classification from Section 2. layouts in 2D [HMM00], Cone trees [RMC91] in 3D, point
We introduce techniques for static and time-varying graphs. based trees [SSH09], nature inspired Phyllotrees [NCA06],
In each part, techniques for hierarchies, generic directed and or hyperbolic layouts [LRP95,Mun97,AH98] (see Figure 7).
undirected graphs, and compound graphs are presented. We Most of these classic tree layout algorithms have a linear
discuss different ways of visual graph representations and complexity in time and memory so the layout computation
designs of graph drawings. is scalable. However, the node-link representation by de-
sign leaves significant background space empty and thereby
3.1. Visual Representations of Static Graphs may encounter scalability problems when applied to larger
graphs. For the visualization of node attributes, specialized
The visualization of static graphs has received much atten- techniques for multi-dimensional data visualization such as
tion in the Information Visualization community. The sec- glyphs, radial or parallel plots have been used.
tion start with trees that are simpler than general graphs.
Space filling techniques These techniques try to use the
3.1.1. Trees
full area of the display to present the hierarchy. Instead of
Techniques for displaying trees can be divided into three employing links for representing node relationships, the spa-
main groups: Space filling, node-link based, and hybrid (see tial positions of nodes are employed, using either closeness

submitted to COMPUTER GRAPHICS Forum (1/2011).

6 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

very good (low) aspect ratios but low readability. In or-

der to better distinguish the hierarchical structure, cush-
ion Treemaps [vWvdW99] apply shading of the shapes.
Treemaps that reflect the geographic distribution of the
hierarchical data were presented in [WD08].
• Adjacency In contrast to Treemaps, adjacency-based tech-
niques do not overlap the parent nodes by child nodes,
but represent the node relationships by placing the child
nodes next to their parent nodes. The placement can be in
circular layers, such as in the SunBurst method [SZ00],
(a) Phyllotrees (b) Point-based tree or on linear layers, yielding so-called icicle plots. The
advantage of this visualization is that the parent nodes
Figure 7: Examples of node-link tree visualizations. are not overlapped by their child nodes and therefore,
c
a) Phyllotrees [NCA06], 2006 IEEE. b) Point-based their attributes can be more easily displayed and analyzed.
c
tree [SSH09], 2009 IEEE. However, this visualization is not as dense as squarified
Treemaps.
• Crossings The crossing method places child nodes across
the parent node, thereby only partially overlapping the
or enclosure. They are mainly applied to visualization of hi-
parent. The Beamtree method [vHvW02] improves over
erarchic partitions of sets of data items, for instance files
the classic Treemaps when the hierarchical structure may
in a file system. Area size can be used to encode quan-
be difficult to visually assess, while still being more space
titative attributes of nodes, such as file size. Additionally,
efficient than the adjacency techniques. The main draw-
color and height can represent additional data attributes. In
back of this technique is that users are unfamiliar with
case more complex additional information needs to be dis-
this approach and that it is often less readable than other
played, specialized data presentations can be placed in the
methods.
child nodes such as icons, parallel coordinate diagrams, etc.
Space-filling techniques can be categorized by the placement
strategy employed into enclosure, adjacency and crossing
(see Figure 8).
• Enclosures These techniques recursively layout child
nodes within the area of their parent nodes. The most
prominent examples are Treemaps – rectangular shapes
recursively subdividing rectangular display space accord-
ing to the underlying hierarchy, introduced by Shneider-
man [Shn92] (so called slice-and-dice algorithm). Vari- (a) Treemap (b) Icicle plot
ants include Voronoi tessellations [BDL05] or bubble lay-
outs [Bed01]. Other types, such as elliptic [OCNF09] or
circular shapes have been proposed, but they do not lead
to fully space filling visualizations.
The main advantage of enclosures is the very good usage
of the available space, as the child nodes do not need extra
space owing to the overlap with the parent nodes. The dis-
advantage is that the overlapping of the parent nodes may
also lead to a more difficult distinction of the hierarchy (c) BeamTree
structure by the user, as it is rather implicitly encoded. For
Treemaps, several advanced layout techniques have been Figure 8: Three types of space filling hierarchy visualization
developed including ordered (i.e., pivot-based) [BSW02], techniques. a) Enclosure: Cushion Treemap [vWvdW99],
squarified [BHvW99], and spiral [TS07] Treemap lay- c
1999 IEEE. b) Adjacency: Icicle plot [TS08a], 2008 c
outs. For example, squarified Treemaps aim at generat- IEEE. c) Crossing: BeamTrees [vHvW02], 2002 c IEEE.
ing subrectangles of square-like aspect ratios, supporting
easier comparison of sizes and presentation of additional
diagrams or other elements within the rectangles. Accord-
ing to Tu and Shen [TS07], the slice-and-dice algorithm Hybrid approaches These approaches combine node-link
leads to high aspect ratios with good readability. Strip, diagrams with Treemaps: a part of the hierarchy is displayed
pivot-based and spiral techniques have medium aspect ra- in a Treemap and the rest as a node-link diagram (see Fig-
tios with medium readability. Squarified Treemaps have ure 6c). They present the data in a flexible space-efficient

submitted to COMPUTER GRAPHICS Forum (1/2011).

 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 7

way while still clearly presenting the data structure and em- edge length is homogeneous, and in general, graph substruc-
phasizing the content. The most prominent representative are tures are easily recognizable. This problem is intensively
“elastic hierarchies” [ZMC05]. In connection to interactive studied in the graph drawing community. Given these aes-
determination of the type of visual metaphor used for each thetic goals and constraints, the aim is to find algorithms that
part of the hierarchy, this technique allows for flexible anal- efficiently provide good solutions.
ysis of the data using advantages of both representations.
Note that a specific group of graphs are graphs with ge-
ographic reference, such as transportation graphs. In this
3.1.2. Directed and Undirected Graphs case, the nodes and possibly also edges of the graph have
Techniques for displaying general graphs can be divided into an inherent geographic location, which needs to be taken
three main groups: node-link based, matrix-based, and hy- into consideration in their graphic presentation. Therefore,
brid (see Figure 9). We discuss these in more detail be- a specific graph layout algorithm is not needed for determin-
low. In addition, there are specialized graph drawing tech- ing the position of each node on the screen. However, the
niques, which use new graph visualization techniques. Two fixed node position may exacerbate graph readability prob-
main examples are: graph splatting and graph maps. The first lems, such as crossings and long edges. Visualization of ge-
one forms graphs as two-dimensional scalar fields [vLdL03]. ographic data is a special research field, which we do not
The second one visualizes graphs as maps [GHK10], where address here in detail.
the relationships between nodes are represented as adjacency
When no position is inherently associated with vertices,
between neighboring areas (nodes). Both approaches create
a graph layout algorithm is required. The graph layout re-
an approximate representation of a graph.
search field is very large, and an extensive survey of pro-
A comparison of node-link and matrix techniques is pre- posed techniques is beyond the scope of this report. The lat-
sented by Ghoniem et al. [GFC04]. According to the study, est survey from Herman et al. dates from 2000 [HMM00]
the advantages of node-link diagrams are their intuitiveness, and several new algorithms have appeared since then. The
compactness, and better suitability for path following tasks. related work part in [AAM07,MM08] as well as the compar-
They are more effective for smaller and sparse graphs. The ison in [HJ07] nicely summarize many current techniques. In
Matrix representation inherently does not have edge cross- our report, we classify the techniques according to the type
ings and node overlapping problems, and is thereby suitable of node placement.
also for dense graphs. When using appropriate node order-
ing, they can easily reveal dense substructures in the graph. • Force-based layouts These techniques are based on a sim-
However, they also suffer from scalability in limited display ulation of mechanical laws by assigning repulsive forces
spaces, especially for very large graphs. In visual graph anal- between nodes and attraction forces between endpoints
ysis, graph layout and matrix ordering influence the effec- of edges. Several forces have been described in the lit-
tiveness of these representations. These issues are therefore erature to achieve different properties of the layout. The
in the core of graph visualization research. seminal work of Eades uses an electric force between
charged particles to model node repulsion and spring
forces between the link endpoints to model edge attrac-
tion [Ead84]. Fruchterman and Reingold [FR91] have
then improved the distribution of nodes by adaptation
of the force models and Noack has further improved it
with a more flexible set of force functions to achieve ei-
ther a good space density or a good clustering of nodes.
Kamada and Kawai [KK89] try to layout nodes such as
(a) Node-link (b) Adjacency (c) Combination the Euclidean distance between the nodes is proportional
diagram matrix to the graph-theoretical distance. This family of layouts,
diagram however, does not scale well to graphs of thousands of
nodes or more, due to their complexity. Therefore, im-
Figure 9: Three types of general graph visualization tech- provements have been proposed. For instance, faster cal-
niques: a) Node-link diagram, b) adjacency matrix, c) hy- culation of forces using an efficient GPU implementa-
c
brid. From [HFM07], 2007 IEEE. tion [GHGH09], or using heuristics [FLM95].
• Constraint-based layouts This family of layouts extends
the force-directed approach with constraints on node po-
Node-link representations The main challenge is the lay- sition. These constraints include horizontal and vertical
out (i.e., the placement of the nodes) so that graph read- alignment of nodes, non-overlapping nodes, edge direc-
ability and certain notions of graph aesthetics are supported tion or closeness of grouped nodes [DMW09a]. An ex-
(see Figure 10). Typical requirements include that the nodes ample are orthogonal layouts, where the edges are only
do not overlap, the number of edge crossings is minimized, composed of straight vertical and horizontal lines. These

submitted to COMPUTER GRAPHICS Forum (1/2011).

8 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

layouts can be supported also by user interaction (see erties associated with graph vertices can be used directly
also Section 4). Example works from this category in- to specify the position of these vertices, as with scatter-
clude [DMS∗ 08, DMW09a, DMW09b]. This family of plots [SA06, BCD∗ 10]: the layout computation is then
layouts greatly improves the power of expression at the straightforward and very fast.
cost of slightly longer execution time.
• Multi-scale approaches These techniques rely on a hi-
erarchical decomposition of a graph into simpler nested
sub-graphs. They first layout the coarser graph and then
include more nodes level by level. Exemplary works
include [GK01, KCH02, HJ05, FT07, MM08] (see Fig-
ure 10a). These methods are typically much faster than
(a) Multi-level graph layouts
traditional force-directed methods. They can be differenti-
ated according to the technique used for creating the node
hierarchy, and the layout of the resulting layers. For ex-
ample, [MM08] employs node clustering and subsequent
positioning of the nodes along space filling curves.
• Layered layouts These approaches, also called “hierarchic
layouts”, place nodes of the graph on parallel horizon-
tal layers [GKNV93]. They are mainly used for directed
graphs and are based on the Sugiyama approach [STT81].
It works in four phases: (1) cycle removal, (2) assign- (b) Dig-Cola layout
ment of nodes to layers, (3) reduction of edge crossings
and (4) assignment of coordinates to nodes. Improve- Figure 10: Graph layout examples. a) A comparison
ments to these layouts, specifically for cyclic graphs, po- of multi-level graph layouts GRIP, FM3 and Topolay-
sition all nodes of a cycle within one level; examples c
out [AAM07]. 2007 IEEE. b) Layered layout of cyclic di-
include the Dig-Cola layout [DK05] and Cyclic Level- c
rected graph [DK05]. 2005 IEEE.
ing [BBBL09] (see Figure 10b). This algorithm and its
variants are quite fast in practice and standard implemen-
tations such as [GKNV93] can easily layout several thou-
sands of nodes in seconds.
• Non-Standard Layouts Other approaches exist that com-
bine the previous techniques or use completely alterna-
tive approaches to graph layouts. Projection of a node
layout from high-dimensional to two-dimensional space
has been proposed in [HK02]; although it is very fast in
practice, the quality of the layout is very sensitive to the
structure of the graph. For example, it is very effective for
meshes and not effective at all for trees. LGL [ADWM04]
first simplifies the graph by computing a spanning-tree; it
then computes the layout iteratively in depth order using
a force-directed layout. LGL is able to scale to very large
graphs (billions of vertices) thanks to the initial decompo- Figure 11: Graph visualization using data seman-
sition. It is very effective for quasi-trees but has not been c
tics [SA06]. 2006 IEEE.
thoroughly studied for other kinds of graphs; its results are
very sensitive to the spanning-tree computation: choos-
ing different spanning trees will results in quite different Comparison of graph layouts A recent comparison of
layouts for the same graph. The ISOM method [Mey98] the readability of graph layouts using eye-tracking [Hua07,
applies the Self-Organizing Map algorithm [Koh01] for PSD09] has shown that force directed layouts outperform or-
finding a suitable graph layout. As an alternative to costly thogonal and layered layouts on various user tasks. Another
layout computation, a graph layout visualization based on comparison of advantages and disadvantages of numerous
the semantics of the graph (on node labels) was presented current layouts was published by Hachul and Jünger [HJ07].
in [SA06]. Semantically identical nodes (e.g., with the They compare the graph drawing outputs according to var-
same labels) can be placed in boxes using standard lay- ious criteria finding that the HDE layout [HK02] is very
out algorithms (e.g., force-directed) (see Figure 11) or in fast but frequently produces layouts with many overlap-
layers using their importance for assigning the position ping edges. In contrast, FM3 [HJ05] creates pleasing lay-
within layers [GOB∗ 10]. Furthermore, attributes or prop- outs in reasonable time. Both algorithms together with

submitted to COMPUTER GRAPHICS Forum (1/2011).

 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 9

GRIP [GK01] scale well with graph size. A comparison of tiling and alternate-bisection. The polyomino algorithm of
user-produced vs. automatically generated layouts [vHR08, Freivalds et al. [FDK02] uses a special representation of the
DLF∗ 09] found also that the results of physics-based algo- graph objects, which substantially reduces the unused dis-
rithms, such as force-directed layouts, were preferred by the play space in comparison to rectangular shapes. Goehlsdorf
users. et al. [GKS07] introduce new quality measures to evaluate a
two-dimensional placement which yields more compact lay-
Design of graph drawing The above mentioned techniques outs than the previously mentioned approaches.
cover graph layout. In addition to specific layouts, occlu-
sion and readability of the display can be improved by edge- Matrix Representation These techniques visualize the ad-
bundling [Hol06, CZQ∗ 08, TE10, LBA10] (see Figure 12) jacency matrix of a given graph, where edge attributes are
and the removal of node overlap [GH09, IAG∗ 09]. Draw- encoded in the matrix cells. They can display both directed
ing of node-link diagrams also includes a suitable design and undirected graphs, where the latter leads to a symmet-
of edge and node drawing primitives. For directed graphs, ric matrix. The advantage of this representation with respect
the representation of edge directions is of importance. There to the node-link representation is the non-overlapping dis-
are multiple design possibilities including usage of arrows, play of graph edges, and the readability of the graph espe-
color transitions (from color A to color B), thickness tran- cially for larger and denser graphs. The disadvantage is an
sitions (from thick to narrow), curves, and animated tex- increased difficulty for users to follow paths, and a possi-
tures [TK08, HvW09, BBG∗ 09]. These options may also be ble unfamiliarity of matrices to the users. In a matrix visu-
combined. A comparison of graph drawing different ways alization, the ordering of rows/columns plays an important
to represent edges was presented in [HvW09]. It shows that role: similar to layout for the node-link representation. Dif-
arrows, although popular and widely used, do not perform ferent strategies to order the matrix can be employed (see
as well as color and thickness transitions. Graph nodes and Figure 13). Prespicuous reordering can reveal clusters in the
edges often have associated attributes that are included in the graph and other patterns. For a discussion of these, we re-
analysis. This study did not concentrate on attributed edges. fer to [MML07,DPS02,HF06,EDG∗ 08]. Although matrices
For such edge attributes, in particular edge weight, coloring are suitable for larger graphs, they also suffer from scala-
of edges or edge thickness can be employed. For the visu- bility issues as they use linear order of nodes along the ma-
alization of node attributes, a visualization of multivariate trix rows/columns. Therefore, interaction techniques and ag-
data items (e.g., glyphs or radial plots) is employed. Various gregated displays have been proposed [vH03, AvH04, HF06,
possibilities of graph designs can be found in [Kre09]. EDG∗ 08, vHSD09] (see also Sections 4 and 5).

(a) Original graph (b) Edge bundling

(a) HDE matrix ordering (b) NNTSP matrix ordering
Figure 12: The use of edge bundling for improving
graph readability. a) original graph b) graph with edge Figure 13: Examples of matrix reordering on graph presen-
c
bundling. [Hol06], 2006 IEEE. tation. a) Using HDE algorithm. b) Using NNTSP reorder-
ing. From [EDG∗ 08], 2008
c IEEE.

Visualization of multiple graph connected components

For the visualization of multiple components, first a lay- Combination of matrix and node-link approach Tech-
out for each individual connected component is calculated niques using a combination of the two previous approaches
and then a specific placement of these components on aim at overcoming their limitations by focusing on their
the screen is performed. The most widely used placement strengths. Three main approaches exist (see Figure 14).
method is called packing. It lays out the components so • Multiple synchronized views These techniques link the
that they do not overlap and are space efficient. Dogru- matrix and node-link representation [HF06]. Both views
soz [Dog02] compares several two-dimensional packing al- show the same data and are synchronized during explo-
gorithms for graphs which use representation of graphs ration. Thereby, the user can concentrate on whatever
by their bounding rectangles. They include strip packing, view is more suitable for the current task.

submitted to COMPUTER GRAPHICS Forum (1/2011).

10 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

• Matrix with link overlay The Matlink [HF07] approach

enhances matrix visualization with links at the border of
the matrix (connecting the nodes). Using link highlight-
ing, the paths can be easily spotted in the Matlink view
and at the same time, the advantages of the matrix repre-
sentation are retained.
• Partial matrix and node-link representation There are two
main approaches. Firstly, Nodetrix [HFM07] combines
both representations in one view, where node-link dia- (a) Multiple linked views showing the same data using different rep-
resentations
grams display the overall graph structure of the network,
and adjacency matrices show communities. The work also
discusses three ways of link display for this setting: aggre-
gated links, underlying links, and underlying links with
full size (see Figure 15). These forms can be also used
for attributed links. Secondly, layered graphs (directed
acyclic graphs) can be represented by so-called “quilts”.
They arrange nodes in a matrix-like form and connect
them with orthogonal edges. In this way, a clear view of
the graph is created [WBS∗ 08, BDF∗ 10].

3.1.3. Compound Graphs

Literature on visualization of graphs with hierarchic struc-
ture is relatively rare. We identify three main approaches.

Node-link graph visualization techniques These use

node-link diagrams for the lowest hierarchy level and
then use “bubbles” (enclosures) for various hierarchy lev-
els. Examples include TugGraph [AMA09] and Grouse-
(b) Links connected to the matrix view highlighting paths between
Flocks [AMA08]. The advantage of this method is its in-
nodes
tuitiveness. However, for large graphs with many links,
this view gets easily overcrowded (see Figure 16 a). The
edge overplotting problem can be partially solved by edge
bundling [Hol06] (see Figure 12). Alternatively, only links
between merged nodes can be drawn (see Figure 16 c).

Treemap-based A treemap visualization of the node hierar-

chy uses overlaid links between nodes [FWD∗ 03] (see Fig-
ure 16b). This approach may suffer from strong overplotting
in case of many links between nodes of the hierarchy. There-
fore, edge bundling is advised to improve the readability
of the display [Hol06] (see Figure 12). Similarly, also one-
dimensional treemaps with links between nodes, so called
ArcTrees [BDJ05] can be employed (see Figure 16d), but
these do not scale well for large hierarchies.
(c) Node-link and matrix combined showing dense areas as matrices
Matrix view with links These visualizations combine the (avoiding edge crossings)
generic node relationship visualization with a tree-based vi-
sualization of the hierarchic node relationships. This is an Figure 14: Examples of combined matrix and node
analogy to MatLink [HF07]. This view is very clear, how- link graph visualization techniques. a) Multiple linked
c
views [HF06]. 2006 IEEE. b) Links connected to the
ever, it may be difficult to understand the compound rela-
c
matrix view [HF07]. 2007 Springer-Verlag Berlin Hei-
tionships between nodes (see Figure 16e).
delberg. c) Node-link and matrix combined - part and
c
part [HFM07]. 2007 IEEE.
3.2. Visual Representation of Dynamic Graphs
In this section, we discuss two categories of visual display
of the time changes on graph elements: Using animation and

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 11

fect also data relationships. Please note that visual analy-

sis of changes in dynamic graphs is related to comparing
graphs. Graph comparison is discussed in Section 5.2.

3.2.1. Trees
(a) Aggregated (b) Underlying (c) Underlying
links links links with full
size

Figure 15: Three ways of link visualization in a combined

node-link and matrix data representation using the NodeTrix
c
approach [HFM07]. 2007 IEEE.

(a) Time line tree

Figure 16: Visualization techniques for compound

c
graphs [Hol06], 2006 IEEE. (a) Node-link visualization
with grouped nodes in “bubbles”. (b) Links overlaying
a treemap visualization. (c) Compound drawing using
enclosures and links between merged nodes. (d) ArcTrees - (b) Treemap with time series
links overlaying a 1D treemap [BDJ05]. (e) A matrix view
for showing relations between entities linked with tree view
of the nodes as in MatLink approach [HF07].

using static displays. Animated displays usually employ or

(c) Animated circular icicle plot
enhance static visualization techniques such as presented in
Section 3.1. Animation is a natural way of conveying the Figure 17: Visualization of time-dependent trees. (a) Time
change of the data over time. However, its effectiveness is c
line tree [BBD08], 2008 ACM. (b) Time series in the
limited by human perception capabilities. Usually, users are treemap nodes [DHKS05], 2005c IEEE. (c) Animated hi-
only able to recognize and remember larger changes in the c
erarchic circlular plots [TS08a], 2008 IEEE.
data. Therefore, highlighting of graph changes is used. It al-
lows for more effective spotting of differences between two
successive time points [APP10]. The static view is preferred For the visualization of dynamic trees with only data
for more detailed analysis of data changes. Static views that attribute changes, either treemaps with time series in the
also incorporate the time-dimension of the data are more leaf nodes [DHKS05, SKM06] or the so called Timeline
complex. In the following, we categorize the visualization Trees [BBD08] can be used (see Figure 17 a and b). Time-
techniques according to the type of data changes captured line trees show the hierarchy on one side and the time se-
into those that affect only data attributes, and those that af- quences on the other side of the view. The treemap repre-

submitted to COMPUTER GRAPHICS Forum (1/2011).

12 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

sentation directly shows the hierarchic structure and time-

variation in one combined view. This allows for an easy
comparison of the time-developments across the hierar-
chy. However, the comparison is affected by different node
sizes and difficult for small nodes. Therefore, a specific
treemap layout preserving the aspect ratio has been devel-
oped [DHKS05, SKM06]. Timeline Trees assign the same
space to all nodes. The vertical positioning of time lines al-
lows for very good comparison of the values at the same time
points. The separation of the time dimension from the hierar-
chic structure, however, complicates the comparison of tree
branches.
For visualization of dynamic data with structural changes,
animated views are used. Card et al [CSP∗ 06] have used
and extension of DOI Trees [CN02, HC04] to visualize the
changes of an administration over time; A time-slider is used
(a) Node-link diagram with time series in
to control the visualized time-span. Animated graphs (see
nodes
Section 3.2.2) can be employed in general. In particular,
the layouts based on the Sugiyama approach [GBPD04] are
suitable. Alternatively, animated treemaps [GF01, TS07] or
icicle/circular plots [TS08a] can be used (see Figure 17 c).
When choosing the graph layout, the layout stability needs
to be taken into consideration. E.g., in the treemap represen-
tations, the spiral layout [TS07] achieves a high continuity
with high stability of the layout. Strip and pivot-by-middle
layouts have also been shown to have higher layout suitabil-
ity [BSW02]. All these layouts are preferable in spite of their
higher aspect ratios in comparison to the squarified treemap.
(b) Animated node-link diagram
Furthermore, dynamic Voronoi treemaps [SFL10] offer both
good aspect ratios and stable layouts for displaying dynamic Figure 18: Visualization of time dependent graphs. (a)
data. Alternatively, Tu and Shen [TS07] propose also static c
Time series in nodes [SLN05], 2005 IEEE. (b) Animated
comparison of two time points in a treemap visualization c
graphs [FT08], 2008 IEEE.
(called contrast treemap).

3.2.2. Directed and Undirected Graphs

GPU, thereby accelerating the layout computation (see Fig-
For attribute changes only, techniques for visualization of ure 18b).
static graphs can be combined with visualizations of individ-
Instead of animation, Brandes and Corman [BC03]
ual time dependent data items (e.g., color charts [SLN05])
use the third dimension to show the evolution on time.
are used (see Figure 18a). The advantage of this approach is
GraphDice [BCD∗ 10] uses interaction to switch between
the large number of the available graph layouts.
projections where time can be mapped to one dimension.
In case of structural changes, time-dependent graph lay-
outs (animated graphs) need to be employed [CBTT95, 3.2.3. Compound Graphs
Nor96, DGK01, EHK∗ 03, KG06]. In animated graph visu-
There are only few techniques that visualize time-varying
alization (in analogy to animated tree visualization), a stable
compound graphs. They employ either animation or static
graph layout, which changes minimally, is of essence. A sta-
data representations.
ble graph layout preserves the mental map of the user. It en-
ables the user to follow changes on the screen [ELMS91, Kumar et al. [KG06] present a specific layout for anima-
DGK01] and thereby it facilitates the analysis of graph tion of a node-link diagram with transparent “bubbles” for
changes. In laying out dynamic graphs, there is a large dif- the hierarchic grouping of nodes (see Figure 19a). Frishman
ference between strategies for drawing graphs with known and Tal [FT04] present a layout which focuses on maintain-
histories and those that need to be adjusted in real-time de- ing the clustered structure during the animation. The groups
pending on new data streams. A paper of Frishman and of nodes are displayed using bounding boxes around the
Tal [FT08] addresses this particular issue by proposing an groups. Reitz et al. [RPD09] use dynamic graph layouts for
online algorithm for dynamic layout implemented on the showing areas of interest in dynamic compound graphs.

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 13

A static approach to visualization of dynamic compound

digraphs using TimeArcTrees was presented by Greilich
et al. [GBD09] (see Figure 19b). They show a sequence
of node-link diagrams with horizontal node alignment in
(a) Animated compound graphs a single view, thereby supporting their direct comparison.
TimeRadarTrees [BD08] use radial tree layouts for the hier-
archy and a sequence of circle segments for representation of
the temporal change of the structure (edges) of the Digraph
(see Figure 19c). This view easily gets complex for larger
graphs.

4. User Interaction in Graph Visualization

Interaction helps users solving tasks connected to explo-
ration of graphs. These tasks can be of different nature such
as topology-based or attribute-based [LPS∗ 06]. Topology-
based tasks include finding adjacent nodes, or determining
connections between nodes. Attribute-based tasks include,
(b) TimeArcTrees
e.g., searching for nodes with specific values, and finding
edges of certain types. For each task, one or more interac-
tion techniques can be employed. Standard interaction tech-
niques such as zooming, panning, or brushing and link-
ing [CMS99, War00] are commonly used in graph visual-
ization. In addition, specialized techniques have been devel-
oped for interactive visual graph navigation and exploration.

Interaction and exploration are deeply inter-related. Some

graph analysis systems such as Pajek [dMB05] claim to sup-
port exploratory graph analysis by chaining complex op-
erations on graphs without showing the intermediary re-
sults. However, Ahlberg et al. describe interactions and more
specifically dynamic queries [AWS92] as required to truly
achieve exploration. The main reason is cognitive: exploring
requires several hypothesis to be maintained in short-term
memory which is very limited in capacity. Planning complex
operations without feedback or using a textual syntax con-
(c) TimeRadarTrees sumes all the short-term memory and exploration becomes
impossible from short-term memory alone. Therefore, pro-
Figure 19: Visualization of time varying compound graphs.
viding interactions with immediate feedback for the most
c
(a) Animated graphs by Kumar et al. [KG06], 2006 IEEE.
frequent operations supports exploration. Other less frequent
c
(b) TimeArcTrees [GBD09], 2009 held by the authors. (c)
operation could still be done using more complex mecha-
c
TimeRadarTrees [BD08], 2008 held by the authors.
nisms, as explained in the next section on graph analysis.
The categorization of interaction techniques can be based
on various criteria such as task, user intention [YKSJ07] or
user action [EF09]. These criteria are interrelated. For exam-
ple, one task may include performing several actions, or one
task may correspond to several user intentions. Moreover,
one user intention can be achieved by several user actions
or, vice versa, an action can suit several intentions.
We categorize interaction techniques according to stages
in the Information Visualization reference model of Card
at al. [CR98, CMS99] and user actions.The reference model
has three stages: data, visual form (a.k.a visual abstraction)
and view. The classification criterion is whether the user ac-
tion affects the data (the selection of the displayed data or

submitted to COMPUTER GRAPHICS Forum (1/2011).

14 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

the data values), the visual display of the data (visual pa- Lenses [EDF10] dynamically adapt the color range of
rameters or visual representation), or the view. Data, visu- items inside the focus region to better use the screen color
alization and view manipulation can be used for interac- range when mapping values with a very large dynamic to
tive data exploration and navigation. This categorization fol- the color of nodes or matrix cells.
lows the idea of Elmqvist and Fekete [EF09] and Bertini and • Multiple foci Multiple foci distort several view areas at
Lalanne [BL09]. Please note that these three types of inter- the same time. It is useful for comparing various parts of
action are sometimes closely connected. For example, data the display or focusing on several items that are spread
manipulation may automatically lead to changes of visual across the view. In node-link diagrams either magnifica-
parameters (e.g., data filtering can influence the graph lay- tion of the areas of interest [SZG∗ 96,TS99] or space fold-
out, or zooming can be combined with data filtering forming ing (shrinking of area out of focus) can be used [MGT∗ 03,
a type of semantic zooming). ERHF09] (see Figure 28 bottom right). For treemaps, the
so-called balloon focus can be used for enlarging multi-
4.1. View Interaction ple items in a treemap [TS08b]. This approach keeps the
form of other areas keeping relative position of items un-
Panning and Zooming Panning and zooming allow to nav- changed (see Figure 22).
igate in any direction and change the zoom-level in the view.
For node-link diagrams, a specific type of panning (guided
panning) has been proposed. It allows to navigate along
edges of a selected node and thereby to explore the structure
of the graph. It can be combined with automatic zooming
on the edge and distortion of end-node position closer to the
currently selected node [MCH∗ 09].

Magic Lenses Owing to the limited display space, show-

ing the whole data set may lead to strong overplot-
ting or very small (up to, unreadable) data items. Magic
Lenses [BSP∗ 93], including distortion techniques change
(a) Original view (b) Edge lens
the representation or allocate more space to items in focused
areas and thereby, improve the readability of the data of in- Figure 20: Example of edge lens interaction. (a) Origi-
terest. They are used both for node-link and space filling nal view without lens. (b) Using edge lens From [WCG03],
graph visualization techniques. The changes can concentrate c
2003 IEEE.
either on one area or on multiple areas of the screen. For ge-
ometric changes, the technique is called fisheye views. Inter-
active selection of the focus area helps to explore different
parts of the data in more detail.
• Single focus Graphical fisheye views were introduced
in [SB92]. So called edge lenses resolve strong overlaps
of edges in the view. They displace the edges to a larger
area [WCG03] (see Figure 20). This approach is espe-
cially useful for geographic-based graphs, where node
repositioning is not desired and therefore, cannot help to (a) Original view (b) Local edge lens
solve edge overlap. Another approach uses filtering of in-
teresting edges in a specified area, or moving neighbor
nodes closer to a selected node relying on the graph struc-
ture [MCH∗ 09]. This type of node position change can
be combined with geometric view distortion [TAvHS06]
(see Figure 21). In node-link visualization of hierarchies,
a degree-of-interest function can be used for allocating
more area to more interesting parts of the tree, e.g., in (c) Bring neighbors lens (d) Composite lens
DOITrees [CN02, HC04].
Figure 21: Examples of different types of edge lens in-
None-geometric magic lenses include Excentric Labels
teractions. (a) Original view without lens. (b) Using lo-
and Color Lenses. Excentric Labels [FP99, BRL09] show
cal edge lens. (c) Using bring neighbors lens. (d) Using
labels or other statistics for items contained in dense
composite lens which combines (b), (c) and Fisheye lens.
focus regions (nodes or matrix cells). The informa-
c
From [TAvHS06], 2006 IEEE.
tion is displayed outside the focus region with connec-
tors linking the nodes/cells to their related label. Color

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 15

4.2. Visual Abstraction Interaction

In these approaches, the change of the visual presentation of
the data concerns adjusting the type of visual presentation
and its parameters.
Most of the graph visualization system provide standard
dialog boxes and widgets to change the visual abstraction
parameters, including the layout technique and its various
parameters. Currently, very few systems allow the interac-
tive manipulation of layout parameters, except using indi-
rect manipulation such as sliders, list boxes, radio buttons
(a) Original view
and check boxes. Rich visualization systems provide a large
number of these indirect manipulation widgets which use an
important amount of the screen real-estate and force users to
search for the right widget by reading their labels and trying
to make sense of them, which can be quite long and tedious.
This is why several research work is devoted to providing
more direct mechanism to change the parameters.

4.2.1. Changes of Visual Parameters

These techniques affect the parameters of the visual presen-
tation. They include highlighting of items and other tech-
niques. (b) Balloon focus

Highlighting The emphasis of interesting items is a stan- Figure 22: Multiple foci in a treemap. (a) Original view. (b)
dard interaction technique. Recently, new techniques for c
Using balloon focus. From [TS08b], 2008 IEEE.
highlighting a node and its neighborhood using hotbox and
lasso selections were presented in [MJ09].

Brushing & Linking Multiple coordinated views are used When concentrating on user-defined changes to graph lay-
to show the data from different perspectives. In these views, outs, an approach to easy selection and layout change of
changes in one visualization (e.g., highlighting) are automat- nodes and subgraphs was presented in [MJ09]. Furthermore,
ically transfered to the other views. For example, a matrix interactive adjustment of the layout constraints was pre-
view coupled to a hierarchical view of the data can be used sented in [DMW09a]. For matrix visualizations, user-driven
to reveal important information in the data [AvH04]. reordering of matrix representation was described in [HF06].

Semantic Zooming Semantic zooming combines zooming

Change of visual representation The change of the type
with an increasing level of detail. In particular, graph ag-
of data presentation, e.g., from a matrix to a node-link dia-
gregation can be used for gaining a coarser view on a large
gram was presented in [ZMC05, HFM07]. This change can
graph. The semantic zooming increases the level of detail by
affect the whole data view [HFM07] (see Figure 23) or only
drilling down to lower levels of aggregation of the original
a part of it [ZMC05, HFM07]. By changing the visual repre-
data [EDG∗ 08, AvH04].
sentation, new insights into the data can be reached. In order
to be able to follow the changes, smooth animations across
4.2.2. Changes of Visual Scheme transitions should be used.
Changes of the visual scheme cover changing of the type of
data visualization either by changing the layout or by chang-
ing the visual mapping. 4.3. Data Interaction
Data-level interaction affects the selection of the data to be
Layout change In node-link diagrams, layout change (ad- displayed, or may change the data values and structure.
justment) affects the positions of the data items on the screen
(see Section 3). It can be performed by changing of the Some operations can be done interactively but general
layout type with automatic recalculation of the new layout, graph analysis system provide more sophisticated mecha-
by manual movement of nodes, or by adjusting the layout nisms including scripting languages or powerful macro fa-
parameters including automatic readjustment of the layout. cilities to perform more complex operations.

submitted to COMPUTER GRAPHICS Forum (1/2011).

16 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

(3) level traversal, where nodes at a certain level are dis-

played; (4) range traversal, where nodes in a range of lev-
els are shown; and (5) unbalanced traversal, where certain
branches of a tree are visible (see Figure 24).

(a) (b)

Figure 23: Transformation of visual representation of a

graph from node-link to matrix view. The picture shows five
c
stages of this process [HFM07], 2007 IEEE.

(c) (d)
4.3.1. Data Filtering
These interaction techniques influence which parts of the
data set are displayed. The data filtering may follow three
paths.

A top down approach This approach starts from the whole

graph and then constrains the part of the data set to be vi-
sualized by filtering according to criteria or by manual data (e)
selection. The disadvantage of this approach is the need to
c
Figure 24: Hierarchy traversal strategies [EF09], 2009
show the whole graph at the beginning, which may require
IEEE. (a) above traversal, (b) below traversal, (c) level
higher computational time for the layout and may lead to
traversal, (d) range traversal. (e) unbalanced traversal.
occlusions owing to the limited screen size. The advantage
is gaining an overview of the graph structure first and then
concentrating on interesting parts. • Navigation based on a degree of interest function These
methods start from a selected node, and next the edges
A bottom up approach This approach starts from one se- and nodes of highest interest are shown [Fur86, vHP09].
lected node [Fur86, AF07, vHP09] and successively shows For the determination of the interesting nodes, a spe-
more nodes/connections on demand. There are two main cific degree of interest (DOI) function is used. De-
methods of choosing the additional nodes/edges to be dis- pending on the specification of the DOI function, var-
played: based on graph structure, or based on a degree- ious graph exploration paths can be followed. These
of-interest function. The advantage of this approach is that DOI functions were used for building specific views on
only the most interesting part of the data set is visualized, trees (DOITrees,SpaceTree) [CN02,HC04,PGB02]. In the
however it is difficult to determine the starting point for work of Furnas [Fur86], the DOI of a node depends on the
the exploration and to define the degree-of-interest function. distance to the node in focus and the a priori interest in
Therefore, we consider these methods in more detail. this node (e.g., according to node importance in the net-
• Navigation based on graph structure. These techniques work, or node properties). Van Ham and Perer [vHP09]
reveal/hide that part of the graph that is determined by extended this function with user interest (UI), which re-
the connections between nodes. In graphs, neighborhood flects the current specific exploratory focus of the user.
traversal shows neighbor nodes of a focus node up to
a certain level [HB05]. For hierarchies, several traversal A middle-out approach This method combines both
methods for have been described in [EF09]. The hierar- bottom-up and top-down approaches. It starts with a coars-
chy traversal methods include: (1) above traversal, where ened graph (middle) and then interactively either reduces or
nodes up to a certain level are shown; (2) below traversal, increases the graph coarsening level by hiding visible nodes
where nodes starting from a selected level are displayed; or showing additional nodes [WMC∗ 09]. For determining

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 17

the middle coarsening level and the next interactive steps, 5.1. Analysis of Graph Structure
graph algorithms are used (see Section 5).
In most user tasks, the analysis of the relationships between
entities in the graph and the assessment of the global graph
4.3.2. Changes of data values
structure plays the key role. These tasks may be effectively
In these approaches, the change of the displayed data set supported by a combination of algorithmic graph analysis
result from direct data value manipulation. Specifically, the and interactive visualization. The algorithmic methods al-
user can change the data values on one level or create/change low, e.g., to calculate node/edge properties, identify clusters
graph aggregations. in the graphs, etc., the results of which are visualized interac-
tively. In the following, we summarize the methods accord-
Graph editing The user can interactively delete or add ing to user tasks starting from more simple to more complex
nodes or edges directly in the visual interface. These graph tasks.
editing actions trigger adjustment of the layout, while still
maintaining the layout style and, where reasonable, the cur- Identification of important nodes In networks, some
rent layout topology. Graph editing affects the structural nodes play a specific role owing to their position within
properties of the graph. In particular, the changes can affect the network. For example, so called hubs and authorities
specific types of subgraphs (so-called motifs). Automatic can be identified and visualized in the network, enabling
identification and highlighting of such structural changes faster analysis of the graph [OPPROG09]. The importance
was presented in [vLGRS09]. of nodes and edges is measured by derived quantities (i.e.,
network metrics) such as centrality-based measures [Fre79]
Interactive graph aggregation For simplification of and ranking-measures [WS03]. Network metrics can help
graphs, graph aggregation is often used. The graph aggre- the analysts to explore networks. Color coding of nodes
gation can be pre-defined, or determined interactively by or edges by metric values, or displaying metrics and net-
the user [HF06, AMA08, AMA09]. For example, Grouse- works in multiple linked views (as lists, scatterplots, or
Flocks [AMA08] allows the user to add and remove aggre- parallel coordinates) are used in this respect. They offer
gated nodes on demand (see Figure 25). This allows for vari- the possibility to interactively chose the metrics of inter-
able views on the graph and its structure. est and to filter/highlighting nodes according to these met-
rics [CJM04, PS06, BCD∗ 10, VMCJ10].

Analysis of connections between two nodes Besides fo-

cusing on single nodes, relations between two nodes can
be analyzed, typically by calculation and highlighting of
shortest paths between the entities. Usually, such analysis
(a) Creating new aggregation node is combined with interactive selection of two entities of in-
terest [HB05, HF07, TK08, GBD09] (see Figure 14b).

(b) Deleting an aggregation node

Figure 25: Interactive editing of a graph hierarchy. a) Cre-

ating a new aggregation node by merging of nodes. b) Delet-
ing an aggregation node, thereby revealing the underlying
merged nodes. From [AMA08], 2008c IEEE.

Figure 26: Interactive graph motif search and visualization.

5. Graph Analysis c
From [vLGRS09], 2009 held by the authors.
Algorithmic graph analysis is beneficial during all stages of
the visual graph analysis process. Relevant techniques allow,
e.g., to reduce a large graph to a smaller graph prior to visu- Analysis of graph substructures In many applications,
alization, to search for specific graph structures of interest, specific types of substructures (i.e., motifs) play an impor-
or to find similarities and dissimilarities for generating com- tant role. For example, in social networks, cliques iden-
parative graph views. In this section, we describe a number tify highly connected communities, or feed-forward motifs
of graph analytical approaches. (substructures in form of a triangle where directed edges

submitted to COMPUTER GRAPHICS Forum (1/2011).

18 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

exist from nodes A to B, A to C and B to C) in bio-

logic networks indicate the functional properties of the net-
work [Sch08]. In order to support the substructure analy-
sis, these motifs can be calculated and visualized in the net-
work [MMO05,SS05,KSS06,vLGRS09,MJW∗ 09] (see Fig-
ure 26). The type of structure can be interactively chosen by
the user in order to support various analytical tasks.

Analysis of graph structure on several aggregation lev-

els User-defined or data-driven graph aggregation can re-
veal relationships between groups of entities in a graph. The
grouping may be based on categoric node attributes [Wat06],
or on a pre-defined node hierarchy [AMA09]. It can also be (a) One-to-one graph matching
user-specified [AMA08], on clustering results based on node
properties ‘ [PS06], or depend on structural properties of the
graph [vLGRS09] (see Figures 5 and 25).

Identification of the impact of graph changes on the

structural properties In time-dependent graphs, the role
of the nodes can change over time, therefore analysis and
visualization of topologic properties (e.g., betweenness cen-
trality) of selected nodes has been proposed [PD08,PRB08].
Additionally, when analyzing user-defined changes (in what-
if-scenarios) the impact of node or edge deletion/addition
on local substructure can be analyzed and highlighted [vL-
GRS09].
(b) One-to-one hierarchy matching
5.2. Graph Comparison
One specifically important analytical task is the examination
of the similarities and differences between multiple graphs,
especially focusing on structural aspects. Usually, structural
differences are in the focus. Such difference may be identi-
fied by the identical node labels in both graphs, or by graph
matching algorithms. After the matching, visualization is
employed to explore the differences [AWW09]. There are
various types of analysis which we describe next. (c) One-to-many graph matching

One-to-one node comparison of two graphs Probably the Figure 27: Visualization of graph comparison. a) One-
most common task in graph comparison is the matching of c
to-one graph matching [CC07], 2007 IEEE. b) One-to-
individual nodes from one graph to individual nodes of the c
one hierarchy matching [HvW08], 2009 held by the au-
second graph. The VisLink visualization approach [CC07] c
thors. c) One-to-many graph matching [GDLP09], 2009
was developed to support this task. It shows both graphs on Springer-Verlag Berlin Heidelberg.
separate planes in 3D, and draws matching links between
corresponding nodes (see Figure 27a). For comparison of hi-
erarchies, a similar approach, based on drawing the two hi-
erarchies in opposite parts of the display and linking of their
leaf nodes was proposed in [HvW08] (see Figure 27b). In
both cases, the visibility of matching links can be increased
by edge bundling.

One-to-many nodes comparison of two graphs One-to-

many nodes comparison concerns correspondence of one
node in one graph to many nodes in another graph. Di Gi-
acomo et al. [GDLP09] developed a system that visual-
izes these one-to-many connections with low overlapping of
links (see Figure 27c).

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 19

(a)

(a)

(b)

(b)

Figure 28: Tree comparison. a) Schema of the tree compar-

ison. b) Example of tree comparison using highlighting of
tree differences. The left view shows the traditional view, the
right view is distorted in order to emphasize important parts
of the tree [MGT∗ 03], 2003
c ACM

(c)
Structural differences between two graphs When ana-
lyzing structural differences between two graphs, analysts Figure 29: Visualization of structural differences between
are often interested in identifying which links or parts of two graphs. a) A schematic illustration of graph difference.
the graphs correspond to or differ from the other one. For b) Visualization of graph differences using network overlap-
the analysis of trees, the TreeJuxtaposer system supports ping [FHK∗ 09], 2009
c IEEE. c) Visualization of graph dif-
to analyze and highlight structural differences between two c
ferences using difference hierarchies [Arc09], 2009 held
trees [MGT∗ 03] (see Figure 28). For general graphs, Fung by the author.
et al. [FHK∗ 09] use both multi-level graph views follow-
ing the VisLink approach [CC07], and overlapping of two
networks with highlighting of common structural parts (see
Figure 29a). Archambault [Arc09] uses graph aggregation
and graph filtering to reveal structural differences between
two graphs (see Figure 29b). 6. Concluding Remarks and Future Challenges

Structural similarity among multiple graphs Structural Research on visual graph analysis deals with the inter-
comparison of multiple graphs is often based on their de- related issues of graph drawing, graph presentation, human-
scription by several graph properties such as graph size, den- computer-interaction, and analytics. This state-of-the-art re-
sity, connectedness etc. (see also Section 2.1). These prop- port represents an encompassing overview and systemati-
erties can be used for exploration of large sets of graphs zation of recent developments in this field. Many advances
[FPSG10], or for the determination of structural similar- have been made on individual parts of visual graph analysis.
ity between graphs. Graph similarity may serve as an in- On the other hand, the surveyed literature discusses many
put for clustering of graphs (grouping similar graphs). Clus- important open challenges, that researchers see in need of
tering helps gaining overview of types of graphs in large work. In the following, we summarize key research chal-
graph databases. Interactive combination of graph cluster- lenges. The discussion of the relevant topics is divided into
ing and visualization of clustering results has been proposed three broad areas: graph visualization and interaction, visual
in [vLGS09] (see Figure 30). analysis systems, and conceptual issues.

submitted to COMPUTER GRAPHICS Forum (1/2011).

20 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

graphs, layout stability and on-line graph drawing are the

main points of interest for the future research. In visual anal-
ysis, the understanding of the graph changes needs to be
supported by stable layouts that preserve the mental map
of the analyst thereby allowing them to follow changes on
the screen [DGK01]. These layouts should be very stable for
minor graph changes and, at the same time, be able to effec-
tively show large graph changes. While a non-trivial chal-
lenge, if successfully supported it may lead to easier spot-
ting of structural changes in the graph and thereby, more ef-
ficient and effective analysis. On-line graph drawing, where
the data stream is unpredictable, poses major challenges in
this respect.
Compound graphs as a combined graph type, including ag-
gregated graphs, represent a complex data type. The main
analytical problem there is the understanding of both types
of connections in a graph, as well as the understanding of
the graph structures on multiple abstraction levels. This is a
very cumbersome task, which can be supported by graph vi-
Figure 30: SOM-based graph clustering for analysis of sualization systems. However, the drawing of such complex
types of graph data space and similarities between graphs graphs is still in its infancy.
c
[vLGS09], 2009 IEEE. In the future, also further graph types such as hypergraphs
[KKS09], or graphs with overlapping sets of nodes [HD10]
may become more prominent in visual graph analysis re-
search.
6.1. Graph Visualization and Interaction
Scalability issues in graph drawing There has been much Graph uncertainty Graph visualization by now mainly
interest in the development of faster layout algorithms that deals with drawing graphs with given data, largely disre-
produce more readable layouts for large graphs, also using garding graph uncertainty. Visualization of uncertain data is
parallel computing, as provided e.g., by current CPUs and a general challenge in visual analytics. As has been shown in
GPUs. It is recognized that using a combination of automatic [GS05], the degree of data certainty affects analytical deci-
graph layout generation and user-oriented, interactive lay- sions. Therefore, it is an important issue in visual graph anal-
out steering, better layouts can be obtained. As graphs get ysis. In graph visualization, various types of uncertainty can
larger, graph filtering and aggregation have been the main be regarded. The uncertainty can relate to the graph struc-
means of graph simplification allowing to draw them. Alter- ture (the existence of nodes and edges between them) and/or
natively, the limited screen space leading to strong overplot- on graph attributes (edge and node attributes). For display-
ting in large graph visualization can be avoided by drawing ing node and edge attribute uncertainty, various methods
graphs on large screens, where specialized layouts can be from multivariate data visualization with uncertainty (see
applied [MGL06]. It can be foreseen that work on more so- e.g., overviews given in [PWL97, THM∗ 05, GS06]) could
phisticated graph layouts revealing the main structures in the be applied. However, their applicability to graph visualiza-
whole graphs, or parts thereof, will continue. In particular, tion needs to be studied. When dealing with structural uncer-
user involvement in the graph layout process involving ana- tainty, there are few dedicated techniques. For example, Can-
lytical expertise of the user is a promising approach and may didTree [LRCP07] uses transparency and color for convey-
lead to easier interpretation of the drawings. ing uncertainty in merged graphs. Therefore, it is expected
From an analytical perspective, also the understanding of that more methods will be developed in the future to address
the meaning of the nodes and edges, besides their global graph uncertainty issues.
structure, is necessary. In particular, the readable/non-
overlapping drawing of nodes, edges and their labels is an Perception issues in graph visualization The understand-
important issue. When displaying graphs with labels, even ing of graph structures in visualization strongly depends on
smaller graphs can easily lead to overcrowded displays. This human perception capabilities. Studies of human perception
topic is gaining more interest in visual analytics research. for graph drawing have recently focused on comparison of
graph understanding using varying graph layouts. In graph
Graph types in graph drawing In recent years, the vari- design, studies on edge visualization have shown that the
ety of considered graph types has increased substantially. In edge design has an influence on the graph reading. These
particular, there has been a large amount of work on draw- various studies have given rise to new problems in graph vi-
ing dynamic and compound graphs. When drawing dynamic sualization, which need to be studied in the future.

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 21

Graph Interaction Techniques In graph exploration, re- complexity, and the variety of the problems, it is foreseeable
cently new interaction techniques for various graph types that the research in this area will need to continue.
have been developed. These techniques increasingly make
use of the structural properties of the graph to interactively Collaborative visual graph analysis For solving complex
navigate in the graph (e.g. in [TS08b, vHP09, TAS09]). This analytical tasks concerning multiple large related data sets, a
tendency supports the analytical purpose of graph visualiza- collaboration of several experts is necessary. Recently, the
tion, as analysts can more easily examine the structural re- development of collaborative visual analysis systems has
lationship between entities in the graph. In the future, this received attention [BMZ∗ 06, Kee06, Ise07]. However, col-
direction can be extended. laborative visual graph analysis is not represented promi-
nently. Therefore, the study of collaborative systems includ-
ing graph data sets would be of advantage. The specifics of
6.2. Visual Analysis Systems graph exploration, in particular, need to be studied.
Visual analysis systems In line with Keim’s visual analyt-
Insight provenance for visual graph analysis In Visual
ics process [KAF∗ 08], modern visual graph analysis sys-
Analytics applications, the analytical processes are often
tems should interactively integrate data pre-processing, in-
long-running and/or distributed. To support the reproducibil-
teractive data visualization, and building and visualizing of
ity, reversibility and automation of these processes, user
data models for gaining knowledge from the data. Many vi-
tracking of the graph interaction steps is necessary. As a
sual analysis techniques already include parts of this pro-
basis for tracking, a taxonomy of graph interaction tech-
cess. However, many of them rely on black box computa-
niques is necessary. The theory of interaction is a general
tions (e.g., automatic graph pre-processing, automatic calcu-
Visual Analytics challenge [TC06]. Although several inter-
lation of graph similarities or cliques). To support the vari-
action taxonomies also for insight provenance have been re-
able hypothesis-insight-driven analytical process, more user
cently introduced [GZ08, HMSA08], their applicability and
involvement in the process should be aimed at. The user
the need for their adaptation to graph analysis needs to be
should have full control of the type of the analysis and its
studied. In return, specific classifications of graph interaction
parameters. As this process includes multiple loops, inter-
techniques could be developed. In this report, we have aimed
active feedback possibilities are necessary. Therefore, inte-
to classify them for gaining a concise overview of the cur-
grated visual analysis systems should include such features.
rent state of the research. This classification, however, may
not be directly applicable to user tracking applications.
Integration of various data types in visual analysis
Graphs as data structures capturing relationships between Applications For analytical purposes, standard graph vi-
entities are part of a larger set of data types examined in sualization and analysis methods need to be adapted to
various applications. Usually, the analysis of graphs is un- the specific needs of the particular application domain.
dertaken in combination with analysis of related data sets, For example, there are specialized systems for visualiza-
or other data sets are transformed into graphs for their anal- tion of bio-chemical structures, shareholding structures, and
ysis [CGK∗ 07, BMGK08]. For analysis of the various data many more. Designing graph visualization systems with
sets as a whole, the sole focus on visual graph analysis (in fast adaptability to various data types, analytical tasks and
particular graph exploration) without taking other relevant application-dependent analytical processes is still a chal-
data into account, is not suitable. In the future, larger inte- lenge. Even within one application, often, the network to be
grated visual analytics systems combining research results analyzed needs to be constructed from heterogeneous data
from several areas are needed. sources, and the focus of interest (attributes of nodes and
edges) varies dynamically. Designing such systems is obvi-
Addressing new analytical tasks With the increasing data ously not trivial.
set sizes and their complexity, new analytical tasks arise.
For example, one such task is the examination of the sim-
6.3. Conceptual Issues
ilarities and differences between graphs. This task builds on
the examination of the structure of one graph as discussed Evaluation Evaluation of usability and user acceptability
above. Lately, several papers about visual graph comparison of the techniques including development of the evaluation
for both trees and general graphs have been published (see methodologies is an important future challenge for the Vi-
Section 5). The comparison can concern only two graphs, sual Analytics research area [KMS∗ 08, TC05, TC06, LK07].
trying to match nodes and edges between them. It can fo- Currently, there is a broad discussion in the Visual Ana-
cus on finding similar graphs for one particular graph from lytics community on the appropriate methodology for the
a large set of graphs. It can concern gaining an overview of evaluation of Visual Analytics and information visualiza-
the types of structures in a large set of graphs. It can con- tion systems. This discussion applies also to the visual anal-
centrate on analyzing the similarities of whole graphs or on ysis of graphs. This challenge is expressed in the words
matching of parts of one graph to other graphs. Owing to its of Plaisant et al. in the introduction to the special issue

submitted to COMPUTER GRAPHICS Forum (1/2011).

22 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

of Computer Graphics and Applications [PGS09] “Assess- theseus-programm.de/). It was also partially sup-
ing VA [Visual Analytics] technology’s effectiveness is chal- ported by the German Research Foundation (DFG) within
lenging because VA tools combine several disparate com- the project Visual Feature Space Analysis as part of the Pri-
ponents, both low and high level, integrated in complex in- ority Program on Scalable Visual Analytics (SPP 1335).
teractive systems used by analysts, emergency responders,
and others. ... Traditional evaluation metrics such as task
completion time, number of errors, or recall and precision References
are insufficient to quantify the utility of VA tools, and new [AAM07] A RCHAMBAULT D., AUBER D., M UNZNER T.: Topo-
research is needed to improve our VA evaluation method- layout: Multilevel graph layout by topological features. IEEE
ology.”. When concentrating on the evaluation of graph vi- Transactions on Visualization and Computer Graphics 13, 2
(2007), 305–317. 7, 8
sualization techniques, several approaches have been pro-
posed, ranging from quantitative to qualitative studies. Con- [ACJM03] AUBER D., C HIRICOTA Y., J OURDAN F., M ELAN -
CON G.: Multiscale visualization of small world networks.
trolled experiments measuring accuracy and duration of user
Proceedings of IEEE Symposium on Information Visualization
tasks have been used, for example, to compare tree visu- (2003), 75–81. 4
alization techniques [Kob04, AK07, ZK08]. An extension
[ADWM04] A DAI A. T., DATE S. V., W IELAND S., M ARCOTTE
of these two main measures, the so-called cognitive load E. M.: LGL: creating a map of protein function with an algo-
measure was used for evaluating general graph visualiza- rithm for visualizing very large biological networks. Journal of
tions [HEH09]. Moreover, eye tracking can be employed Molecular Biology 340, 1 (June 2004), 179–190. 8
for quantitative evaluation, e.g., for comparing graph lay- [AF07] A PPERT C., F EKETE J.-D.: Naviguer dans des grands ar-
outs [Hua07, PSD09]. These controlled studies offer a quan- bres avec controltree. In Proceedings of International Conference
titative comparison across techniques, however often suffer of the Association Francophone d’Interaction Homme-Machine
(2007), pp. 139–142. 16
from the focus only on selected low level tasks. Note that
the formulation of these tasks can influence the comparison [AH98] A NDREWS K., H EIDEGGER H.: Information slices: Vi-
result [ZK08]. A combination of quantitative and qualita- sualising and exploring large hierarchies using cascading, semi-
circular discs. In Late Breaking Hot Topic Paper, Proceedings of
tive study has been performed for comparing graph layouts IEEE Symposium on Information Visualization (1998). 5
produced by both in a manual and in an algorithmic way
[AK07] A NDREWS K., K ASANICKA J.: A comparative study of
[DLF∗ 09]. The subjective user view has been used for rank- four hierarchy browsers using the hierarchical visualisation test-
ing of best layouts. A qualitative view on the effectiveness of ing environment (HVTE). In Proceedings of International Con-
visual analytics techniques can be gained by use case stud- ference Information Visualization (2007), pp. 81–86. 5, 22
ies conducted by domain experts (e.g., in [PS08, MGT∗ 03]). [AMA08] A RCHAMBAULT D., M UNZNER T., AUBER D.:
This method offers insights into the usability of the systems Grouseflocks: Steerable exploration of graph hierarchy space.
in real world scenarios, however does not allow for standard- IEEE Transactions on Visualization and Computer Graphics 14,
ized quantitative comparison of the techniques. The choice 4 (July/August 2008), 900–913. 10, 17, 18
of appropriate evaluation method and its design is still dis- [AMA09] A RCHAMBAULT D., M UNZNER T., AUBER D.: Tug-
cussed in the community. graph: Path-preserving hierarchies for browsing proximity and
paths in graphs. In Proceedings of IEEE Pacific Visualization
Symposium (April 2009), pp. 113–120. 10, 17, 18
Taxonomies and benchmarks The field of visual graph
[APP10] A RCHAMBAULT D., P URCHASE H. C., P INAUD B.:
analysis would profit from more elaborate taxonomies for Difference map readability for dynamic graphs. In Proceedings
tasks, interaction, visualization techniques, measures for of International Symposium on Graph Drawing (2010), p. to ap-
quality, and benchmarks for comparing the new techniques. pear. 11
They would support both the design and development of vi- [Arc09] A RCHAMBAULT D.: Structural differences between two
sual analytic systems and their evaluation. Although several graphs through hierarchies. In Proceedings of Graphics Interface
taxonomies and sample data sets exist, a more broader scope (2009), pp. 87–94. 19
of theory and data aspects is needed owing to the large set of [AvH04] A BELLO J., VAN H AM F.: Matrix zoom: A visual inter-
problems/tasks in visual analysis of graphs. face to semi-external graphs. In Proceedings of IEEE Symposium
on Information Visualization (2004), pp. 183–190. 9, 15
[AWS92] A HLBERG C., W ILLIAMSON C., S HNEIDERMAN B.:
Acknowledgements Dynamic queries for information exploration: an implementation
and evaluation. In Proceedings of the SIGCHI conference on Hu-
We thank Vyara Ivanova, Anna Mitkova, Melanie Görner,
man factors in computing systems (New York, NY, USA, 1992),
and Robert Rehner for their helpful comments and sugges- CHI ’92, ACM, pp. 619–626. 13
tions. We are grateful to the anonymous reviewers for their
[AWW09] A NDREWS K., W OHLFAHRT M., W URZINGER G.:
constructive and useful comments. We thank all authors and Visual graph comparison. In Proceedings of International Con-
copyright holders of the original figures for agreeing to their ference on Information Visualisation (2009), pp. 62–67. 18
reproduction in this paper. This work was partially sup- [BBBL09] BACHMAIER C., B RANDENBURG F. J., B RUNNER
ported by the German Federal Ministry of Economics and W., L OVÁSZ G.: Cyclic leveling of directed graphs. Lecture
Technology within the THESEUS project (http://www. Notes in Computer Science 5417 (2009), 348–359. 8

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 23

[BBD08] B URCH M., B ECK F., D IEHL S.: Timeline trees: visu- [BN01] BARLOW T., N EVILLE P.: A comparison of 2-D visual-
alizing sequences of transactions in information hierarchies. In izations of hierarchies. In Proceedings of IEEE Symposium on
Proceedings of the working conference on Advanced visual inter- Information Visualization (2001), pp. 131–138. 5
faces (2008), pp. 75–82. 11 [BRL09] B ERTINI E., R IGAMONTI M., L ALANNE D.: Extended
[BBD09] B ECK F., B URCH M., D IEHL S.: Towards an aes- excentric labeling. Computer Graphics Forum 28 (2009), 927–
thetic dimensions framework for dynamic graph visualisations. 934. 14
In Information Visualisation, 2009 13th International Conference [BRSG07] B ENNETT C., RYALL J., S PALTEHOLZ L., G OOCH
(July 2009), pp. 592–597. 5 A.: The aesthetics of graph visualization. In Proceedings of
[BBG∗ 09] B LAAS J., B OTHA C., G RUNDY E., J ONES M., Computational Aesthetics in Graphics, Visualization, and Imag-
L ARAMEE R., P OST F.: Smooth graphs for visual exploration ing (2007). 5
of higher-order state transitions. IEEE Transactions on Visual- [BSP∗ 93] B IER E. A., S TONE M. C., P IER K., B UXTON W.,
ization and Computer Graphics 15, 6 (2009), 969–976. 9 D E ROSE T. D.: Toolglass and Magic Lenses: the See-Through
[BC03] B RANDES U., C ORMAN S. R.: Visual unrolling of net- Interface. In Proceedings of the 20th annual conference on Com-
work evolution and the analysis of dynamic discourse. Informa- puter graphics and interactive techniques (New York, NY, USA,
tion Visualization 2 (March 2003), 40–50. 12 1993), SIGGRAPH ’93, ACM, pp. 73–80. 14

[BCD∗ 10] B EZERIANOS A., C HEVALIER F., D RAGICEVIC P., [BSW02] B EDERSON B. B., S HNEIDERMAN B., WATTENBERG
E LMQVIST N., F EKETE J.-D.: Graphdice: A system for explor- M.: Ordered and quantum treemaps: Making effective use of 2D
ing multivariate social networks. Computer Graphics Forum 29, space to display hierarchies. ACM Transactions on Graphics 21,
3 (2010), 863–872. 8, 12, 17 4 (October 2002), 833–854. 6, 12
[CBTT95] C OHEN R. F., BATTISTA G. D., TAMASSIA R., T OL -
[BD08] B URCH M., D IEHL S.: Timeradartrees: Visualizing dy-
LIS I. G.: Dynamic graph drawings: Trees, series-parallel di-
namic compound digraphs. Computer Graphics Forum 27, 3
graphs, and planar st-digraphs. SIAM J. Comput. 24 (October
(2008), 823–830. 13
1995), 970–1001. 12
[BDF∗ 10] B EZERIANOS A., D RAGICEVIC P., F EKETE J.-D., [CC07] C OLLINS C., C ARPENDALE S.: VisLink: Revealing re-
BAE J., WATSON B.: GeneaQuilts: A system for exploring large lationships amongst visualizations. IEEE Transactions on Visu-
genealogies. IEEE TRANSACTIONS ON VISUALIZATION AND alization and Computer Graphics 13, 6 (2007), 1192–1199. 18,
COMPUTER GRAPHICS 16, 6 (2010), 1073–1081. 10 19
[BDJ05] B RODLIE K. W., D UKE D. J., J OY K. I.: Arctrees: [CGK∗ 07] C HANG R., G HONIEM M., KOSARA R., R IBARSKY
Visualizing relations in hierarchical data. In Proceedings of the W., YANG J., S UMA E., Z IEMKIEWICZ C., K ERN D., S UD -
Joint Eurographics and IEEE TCVG Symposium on Visualiza- JIANTO A.: WireVis: Visualization of categorical, time-varying
tion (2005), Neumann P., Schlechtweg S., Carpendale S., (Eds.), data from financial transactions. Proceedings of IEEE Sympo-
pp. 53–60. 10, 11 sium on Visual Analytics Science and Technology (2007), 155–
[BDL05] BALZER M., D EUSSEN O., L EWERENTZ C.: Voronoi 162. 21
treemaps for the visualization of software metrics. In Pro- [CJM04] C HIRICOTA Y., J OURDAN F., M ELANCON G.: Metric-
ceedings of ACM symposium on Software visualization (2005), based network exploration and multiscale scatterplot. In Infor-
pp. 165–172. 6 mation Visualization, 2004. INFOVIS 2004. IEEE Symposium on
[BDL∗ 10] BATAGELJ V., D IDIMO W., L IOTTA G., PALLADINO (0-0 2004), pp. 135 –142. 17
P., PATRIGNANI M.: Visual analysis of large graphs using (x,y)- [CMS99] C ARD S. C., M ACKINLAY J., S HNEIDERMAN B.:
clustering and hybrid visualizations. In Proceedings of IEEE Pa- Readings in Information Visualization: Using Vision to Think.
cific Visualization Symposium (2010), pp. 209–216. 4 Morgan Kaufmann Publishers, 1999. 13
[Bed01] B EDERSON B. B.: Photomesa: a zoomable image [CN02] C ARD S. K., NATION D.: Degree-of-interest trees: a
browser using quantum treemaps and bubblemaps. In Proceed- component of an attention-reactive user interface. In AVI ’02:
ings of ACM symposium on User interface software and technol- Proceedings of the Working Conference on Advanced Visual In-
ogy (2001), pp. 71–80. 6 terfaces (New York, NY, USA, 2002), ACM, pp. 231–245. 12,
14, 16
[BHvW99] B RULS M., H UIZING K., VAN W IJK J.: Squarified
treemaps. In Proceedings of the Joint Eurographics and IEEE [CR98] C HI E. H.-H., R IEDL J. T.: An operator interaction
TCVG Symposium on Visualization (1999), pp. 33–42. 6 framework for visualization systems. In Information Visual-
ization, 1998. Proceedings. IEEE Symposium on (Oct. 1998),
[BL09] B ERTINI E., L ALANNE D.: Surveying the complemen- pp. 63–70. 13
tary role of automatic data analysis and visualization in knowl-
edge discovery. In Proceedings of the ACM SIGKDD Workshop [CSP∗ 06] C ARD S. K., S UN B., P ENDLETON B. A., H EER J.,
on Visual Analytics and Knowledge Discovery (2009), pp. 12–20. B ODNAR J. W.: TimeTree: Exploring time changing hierar-
14 chies. Symposium On Visual Analytics Science And Technology
0 (2006), 3–10. 12
[BMGK08] BARSKY A., M UNZNER T., G ARDY J., K INCAID
R.: Cerebral: Visualizing multiple experimental conditions on [CZQ∗ 08] C UI W., Z HOU H., Q U H., W ONG P. C., L I X.:
a graph with biological context. Visualization and Computer Geometry-based edge clustering for graph visualization. IEEE
Graphics, IEEE Transactions on 14, 6 (Nov.-Dec. 2008), 1253– Transactions on Visualization and Computer Graphics 14, 6
1260. 21 (2008), 1277–1284. 9

[BMZ∗ 06] B RENNAN S., M UELLER K., Z ELINSKY G., R A - [DBETT99] D I BATTISTA G., E ADES P., TAMASSIA R., T OL -
LIS I. G.: Graph Drawing: Algorithms for the Visualization of
MAKRISHNAN I., WARREN D., K AUFMAN A.: Toward a multi-
analyst, collaborative framework for visual analytics. In Proceed- Graphs. Prentice Hall, 1999. 5
ings of IEEE Symposium on Visual Analytics Science and Tech- [DGK01] D IEHL S., G OERG C., K ERREN A.: Preserving the
nology (31 2006-Nov. 2 2006), pp. 129–136. 21 mental map using foresighted layout. In Proceedings of Joint

submitted to COMPUTER GRAPHICS Forum (1/2011).

24 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

Eurographics, IEEE TCVG Symposium on Visualization (2001), [FDK02] F REIVALDS K., D OGRUSÖZ U., K IKUSTS P.: Discon-
pp. 175–184. 12, 20 nected graph layout and the polyomino packing approach. In
Revised Papers from the 9th Int. Symposium on Graph Drawing
[DHKS05] DAYAL U., H AO M., K EIM D., S CHRECK T.: Im-
(London, UK, 2002), Springer-Verlag, pp. 378–391. 9
portance driven visualization layouts for large time-series data.
In Proceedings of IEEE Symposium on Information Visualization [FHK∗ 09] F UNG D. C. Y., H ONG S.-H., KOSCHUTZKI D.,
(2005), pp. 203–210. 11, 12 S CHREIBER F., X U K.: Visual analysis of overlapping biological
networks. In Proceedings of the 2009 13th International Confer-
[Die05] D IESTEL R.: Graph Theory. Springer-Verlag, Heidel- ence Information Visualisation (Washington, DC, USA, 2009),
berg, 2005. 2, 3 IEEE Computer Society, pp. 337–342. 19
[DK05] DWYER T., KOREN Y.: DIG-COLA: Directed graph [FLM95] F RICK A., L UDWIG A., M EHLDAU H.: A fast adap-
layout through constrained energy minimization. In Proceedings tive layout algorithm for undirected graphs. In Proceedings of
of IEEE Symposium on Information Visualization (2005), p. 9. 8 the DIMACS International Workshop on Graph Drawing (1995),
[DLF∗ 09] DWYER T., L EE B., F ISHER D., Q UINN K. I., I SEN - pp. 388–403. 7
BERG P., ROBERTSON G., N ORTH C.: A comparison of user- [FP99] F EKETE J.-D., P LAISANT C.: Excentric labeling: dy-
generated and automatic graph layouts. In IEEE Symposium on namic neighborhood labeling for data visualization. In Proceed-
Information Visualization (2009). 9, 22 ings of the SIGCHI conference on Human factors in computing
[dMB05] DE N OOY W., M RVAR A., BATAGELJ V.: Exploratory systems: the CHI is the limit (New York, NY, USA, 1999), CHI
Social Network Analysis with Pajek. CUP, 2005. 13 ’99, ACM, pp. 512–519. 14
[FPSG10] F REIRE M., P LAISANT C., S HNEIDERMAN B., G OL -
[DMS∗ 08] DWYER T., M ARRIOTT K., S CHREIBER F.,
BECK J.: ManyNets: an interface for multiple network analy-
S TUCKEY P., W OODWARD M., W YBROW M.: Exploration of
sis and visualization. In Proceedings of international conference
networks using overview+detail with constraint-based coopera-
on Human factors in computing systems (New York, NY, USA,
tive layout. IEEE Transactions on Visualization and Computer
2010), ACM, pp. 213–222. 19
Graphics 14, 6 (Nov.-Dec. 2008), 1293–1300. 8
[FR91] F RUCHTERMAN T. M. J., R EINGOLD E. M.: Graph
[DMW09a] DWYER T., M ARRIOTT K., W YBROW M.: Dun-
drawing by force-directed placement. Softw. Pract. Exper. 21,
nart: A constraint-based network diagram authoring tool. Graph
11 (1991), 1129–1164. 7
Drawing 5417 (2009), 420–431. 7, 8, 15
[Fre79] F REEMAN L. C.: Centrality in social networks. Social
[DMW09b] DWYER T., M ARRIOTT K., W YBROW M.: Topol- Networks 1, 3 (1979), 215–239. 17
ogy preserving constrained graph layout. In Revised Papers from
International Symposium on Graph Drawing (2009), pp. 230– [FT04] F RISHMAN Y., TAL A.: Dynamic drawing of clustered
241. 8 graphs. In Proceedings of the IEEE Symposium on Information
Visualization (2004), pp. 191–198. 12
[Dog02] D OGRUSOZ U.: Two-dimensional packing algorithms
for layout of disconnected graphs. Information Sciencies 143, [FT07] F RISHMAN Y., TAL A.: Multi-level graph layout on the
1-4 (2002), 147–158. 9 GPU. IEEE Transactions on Visualization and Computer Graph-
ics 13, 6 (2007), 1310–1319. 8
[DPS02] D ÍAZ J., P ETIT J., S ERNA M.: A survey of graph layout
problems. ACM Computing Surveys 34, 3 (2002), 313–356. 2, 9 [FT08] F RISHMAN Y., TAL A.: Online dynamic graph drawing.
IEEE Transactions on Visualization and Computer Graphics 14,
[Ead84] E ADES P.: A heuristic for graph drawing. Congressus 4 (2008), 727–740. 12
Numerantium 42 (1984), 149–160. 7
[Fur86] F URNAS G. W.: Generalized fisheye views. SIGCHI
[EDF10] E LMQVIST N., D RAGICEVIC P., F EKETE J.-D.: Color Bull. 17, 4 (1986), 16–23. 16
lens: Adaptive color scale optimization for visual exploration.
[FWD∗ 03] F EKETE J.-D., WANG D., DANG N., A RIS A.,
IEEE Transactions on Visualization and Computer Graphics 99,
P LAISANT C.: Overlaying graph links on treemaps. In Pro-
PrePrints (2010). 14
ceedings of IEEE Information Visualization Symposium Posters
[EDG∗ 08] E LMQVIST N., D O T.-N., G OODELL H., H ENRY N., Compendium (2003). 10
F EKETE J.-D.: Zame: Interactive large-scale graph visualization. [GBD09] G REILICH M., B URCH M., D IEHL S.: Visualizing the
In Proceedings of IEEE Pacific Visualization Symposium (2008), evolution of compound digraphs with TimeArcTrees. Computer
pp. 215–222. 4, 9, 15 Graphics Forum 28, 3 (2009), 975–982. 13, 17
[EF09] E LMQVIST N., F EKETE J.-D.: Hierarchical aggregation [GBPD04] G ÖRG C., B IRKE P., P OHL M., D IEHL S.: Dy-
for information visualization: Overview, techniques, and design namic graph drawing of sequences of orthogonal and hierarchical
guidelines. IEEE Transactions on Visualization and Computer graphs. In Graph Drawing (2004), Pach J., (Ed.), vol. 3383 of
Graphics 99 (2009). 13, 14, 16 Lecture Notes in Computer Science, Springer, pp. 228–238. 12
[EHK∗ 03] E RTEN C., H ARDING P. J., KOBOUROV S. G., [GDLP09] G IACOMO E., D IDIMO W., L IOTTA G., PALLADINO
WAMPLER K., Y EE G.: Graphael: Graph animations with evolv- P.: Visual analysis of one-to-many matched graphs. In Revised
ing layouts. In In 11th Symposium on Graph Drawing (2003), Papers from International Symposium on Graph Drawing (2009),
pp. 98–110. 12 pp. 133–144. 18
[ELMS91] E ADES P., L AI W., M ISUE K., S UGIYAMA K.: Pre- [GF01] G HONIEM M., F EKETE J.-D.: Animating treemaps. In
serving the Mental Map of a Diagram. Research Report IIAS- Proceedings of 18th HCIL Symposium - Workshop on Treemap
RR-91-16E, International Institute for Advanced Study of Social Implementations and Applications (2001). 12
Information Science, August 1991. 12
[GFC04] G HONIEM M., F EKETE J.-D., C ASTAGLIOLA P.: A
[ERHF09] E LMQVIST N., R ICHE Y., H ENRY N., F EKETE J.-D.: comparison of the readability of graphs using node-link and
Mélange: Space folding for visual exploration. IEEE Transac- matrix-based representations. In Proceedings of IEEE Sympo-
tions on Visualization and Computer Graphics 99, 1 (2009). 14 sium on Information Visualization (2004), pp. 17–24. 7

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 25

[GH09] G ANSNER E. R., H U Y.: Efficient node overlap removal [HJ05] H ACHUL S., J ÜNGE M.: Drawing large graphs with a
using a proximity stress model. Graph Drawing 5417 (2009), potential-field-based multilevel algorithm. Lecture notes in Com-
206–217. 9 puter Science 4372 (2005), 285–295. 8
[GHGH09] G ODIYAL A., H OBEROCK J., G ARLAND M., H ART [HJ07] H ACHUL S., J ÜNGER M.: Large-graph layout algorithms
J. C.: Rapid multipole graph drawing on the gpu. In Graph at work: An experimental study. Journal of Graph Algorithms
Drawing, Tollis I. G., Patrignani M., (Eds.). Springer-Verlag, and Applications 11, 2 (2007), 234–369. 7, 8
Berlin, Heidelberg, 2009, pp. 90–101. 7
[HK02] H AREL D., KOREN Y.: Graph drawing by high-
[GHK10] G ANSNER E., H U Y., KOBOUROV S.: GMap: Visual- dimensional embedding. In Revised Papers from the 10th In-
izing graphs and clusters as maps. In Proceedings of IEEE Pacific ternational Symposium on Graph Drawing (2002), pp. 207–219.
Visualization Symposium (2010), pp. 210–208. 7 8
[GK01] G AJER P., KOBOUROV S. G.: GRIP: Graph drawing [HMM00] H ERMAN I., M ELANCON G., M ARSHALL M. S.:
with intelligent placement. In Proceedings of the 8th Interna- Graph visualization and navigation in information visualization:
tional Symposium on Graph Drawing (2001), pp. 222–228. 8, A survey. IEEE Transactions on Visualization and Computer
9 Graphics 6, 1 (/2000), 24–43. 2, 3, 5, 7
[GKNV93] G ANSNER E. R., KOUTSOFIOS E., N ORTH S. C., [HMSA08] H EER J., M ACKINLAY J., S TOLTE C., AGRAWALA
VO K.-P.: A technique for drawing directed graphs. IEEE Trans. M.: Graphical histories for visualization: Supporting analysis,
Softw. Eng. 19 (March 1993), 214–230. 8 communication, and evaluation. Visualization and Computer
[GKS07] G OEHLSDORF D., K AUFMANN M., S IEBENHALLER Graphics, IEEE Transactions on 14, 6 (Nov.-Dec. 2008), 1189
M.: Placing connected components of disconnected graphs. In –1196. 21
Proceedings of Asia-Pacific Symposium on Information Visuali- [Hol06] H OLTEN D.: Hierarchical edge bundles: Visualization of
sation (Februar 2007), pp. 101–108. 9 adjacency relations in hierarchical data. IEEE Transactions on
[GOB∗ 10] G RETARSSON B., O’D ONOVAN J., B OSTANDJIEV Visualization and Computer Graphics 12, 5 (2006), 741–748. 9,
S., H ALL C., H ÖLLERER T.: SmallWorlds: Visualizing social 10, 11
recommendations. Computer Graphics Forum 29, 3 (2010), 833– [Hua07] H UANG W.: Using eye tracking to investigate graph lay-
842. 8 out effects. In Proceedings of Asia-Pacific Symposium on Visual-
[GS05] G RIETHE H., S CHUMANN H.: Visualizing uncertainty ization (2007), pp. 97–100. 8, 22
for improved decision making. In Proceedings of 4th Interna- [HvW08] H OLTEN D., VAN W IJK J. J.: Visual comparison of
tional Conference on Perspectives in Business Informatics Re- hierarchically organized data. Computer Graphics Forum 27, 3
search (BIR 2005) (2005). 20 (2008), 759–766. 18
[GS06] G RIETHE H., S CHUMANN H.: The visualization of un-
[HvW09] H OLTEN D., VAN W IJK J. J.: A user study on vi-
certain data: Methods and problems. In Proceedings of Simula-
sualizing directed edges in graphs. In Proceedings of inter-
tion and Visualization (2006), pp. 143–156. 20
national conference on Human factors in computing systems
[GZ08] G OTZ D., Z HOU M.: Characterizing users’ visual an- (2009), pp. 2299–2308. 9
alytic activity for insight provenance. In Proceedings of IEEE
Symposium on Visual Analytics Science and Technology (2008), [IAG∗ 09] I MAMICHI T., A RAHORI Y., G IM J., H ONG S.-H.,
pp. 123–130. 21 NAGAMOCHI H.: Removing node overlaps using multi-sphere
scheme. In Graph Drawing (2009), pp. 296–301. 9
[HB05] H EER J., B OYD D.: Vizster: visualizing online social
networks. In Information Visualization, 2005. INFOVIS 2005. [Ise07] I SENBERG P.: Information visualization in co-located col-
IEEE Symposium on (Oct. 2005), pp. 32–39. 16, 17 laborative environments. In Proceedings of the Grace Hopper
Celebration of Women in Computing, PhD Forum (2007). 21
[HC04] H EER J., C ARD S. K.: Doitrees revisited: scalable,
space-constrained visualization of hierarchical data. In AVI ’04: [JHGH08] J IA Y., H OBEROCK J., G ARLAND M., H ART J.: On
Proceedings of the working conference on Advanced visual inter- the visualization of social and other scale-free networks. IEEE
faces (New York, NY, USA, 2004), ACM, pp. 421–424. 12, 14, Transactions on Visualization and Computer Graphics 14, 6
16 (2008), 1285–1292. 4
[HD10] H ENRY R ICHE N., DWYER T.: Untangling euler di- [JS10] J ÜRGENSMANN S., S CHULZ H.-J.: Poster: A visual sur-
agrams. IEEE TRANSACTIONS ON VISUALIZATION AND vey of tree visualization. In Proceedings of IEEE Information
COMPUTER GRAPHICS 16, 6 (2010), 1090–1099. 20 Visualization (2010). 5
[HEH09] H UANG W., E ADES P., H ONG S.-H.: Measuring ef- [KAF∗ 08] K EIM D., A NDRIENKO G., F EKETE J.-D., G ÖRG
fectiveness of graph visualizations: a cognitive load perspective. C., KOHLHAMMER J., M ELANCON G.: Information Visualiza-
Information Visualization 8 (June 2009), 139–152. 22 tion, vol. 4950 of Lecture Notes in Computer Science. Springer,
2008, ch. Visual Analytics: Definition, Process, and Challenges,
[HF06] H ENRY N., F EKETE J.-D.: MatrixExplorer: a dual- pp. 154–175. 2, 21
representation system to explore social networks. IEEE Transac-
tions on Visualization and Computer Graphics 12, 5 (Sept.-Oct. [KCH02] KOREN Y., C ARMEL L., H AREL D.: ACE: A fast mul-
2006), 677–684. 9, 10, 15, 17 tiscale eigenvectors computation for drawing huge graphs. In
Proceedings of the IEEE Symposium on Information Visualiza-
[HF07] H ENRY N., F EKETE J.-D.: Matlink: Enhanced matrix
tion (2002), p. 137. 8
visualization for analyzing social networks. In Proceedings of
the International Conference on Human-Computer-Interaction – [Kee06] K EEL P.: Collaborative visual analytics: Inferring from
Interact (2007), pp. 288–302. 10, 11, 17 the spatial organization and collaborative use of information. In
Proceedings of IEEE Symposium on Visual Analytics Science and
[HFM07] H ENRY N., F EKETE J.-D., M C G UFFIN M.: NodeTrix:
Technology (31 2006-Nov. 2 2006), pp. 137–144. 21
a hybrid visualization of social networks. IEEE Transactions on
Visualization and Computer Graphics 13, 6 (Nov.-Dec. 2007), [KG06] K UMAR G., G ARLAND M.: Visual exploration of com-
1302–1309. 7, 10, 11, 15, 16 plex time-varying graphs. IEEE Transactions on Visualization

submitted to COMPUTER GRAPHICS Forum (1/2011).

26 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

and Computer Graphics 12, 5 (Sept.-Oct. 2006), 805–812. 12, [MGL06] M UELLER C., G REGOR D., L UMSDAINE A.: Dis-
13 tributed force-directed graph layout and visualization. In Eu-
[KK89] K AMADA T., K AWAI S.: An algorithm for drawing gen- rographics Symposium on Parallel Graphics and Visualization
eral undirected graphs. Information Processing Letters 31, 1 (2006). 20
(1989), 7–15. 7 [MGT∗ 03] M UNZNER T., G UIMBRETIÈRE F., TASIRAN S.,
[KKS09] K AUFMANN M., K REVELD M., S PECKMANN B.: Z HANG L., Z HOU Y.: TreeJuxtaposer: scalable tree comparison
Subdivision drawings of hypergraphs. In Revised Papers from In- using focus+context with guaranteed visibility. In In Proceedings
ternational Symposium on Graph Drawing (2009), pp. 396–407. of ACM SIGGRAPH (2003), pp. 453–462. 14, 19, 22
20 [MJ09] M C G UFFIN M. J., J URISICA I.: Interaction techniques
[KMS∗ 08] K EIM D. A., M ANSMANN F., S CHNEIDEWIND J., for selecting and manipulating subgraphs in network visualiza-
T HOMAS J., Z IEGLER H.: Visual analytics: Scope and chal- tions. IEEE Transactions on Visualization and Computer Graph-
lenges, visual data mining: Theory, techniques and tools for vi- ics 15 (2009), 937–944. 15
sual analytics. Lecture Notes In Computer Science (LNCS) 4404 [MJW∗ 09] M A’ AYAN A., J ENKINS S., W EBB R., B ERGER S.,
(2008), 76–90. 1, 21 P URUSHOTHAMAN S., H USN N. A., P OSNER J., F LORES T.,
[Kob04] KOBSA A.: User experiments with tree visualization sys- I YENGAR R.: SNAVI: Desktop application for analysis and vi-
tems. In Proceedings of the IEEE Symposium on Information sualization of large-scale signaling networks. BMC Systems Bi-
Visualization (2004), pp. 9–16. 5, 22 ology 3, 1 (2009). 4, 18
[Koh01] KOHONEN T.: Self-Organizing Maps, 3rd ed. Springer, [MM08] M UELDER C., M A K.-L.: Rapid graph layout using
Berlin, 2001. 8 space filling curves. Visualization and Computer Graphics, IEEE
Transactions on 14, 6 (Nov.-Dec. 2008), 1301–1308. 7, 8
[Kre09] K REMPEL L.: Network visualization. In Handbook of
Social Network Analysis, Scott J., Carrington P. J., (Eds.). Sage, [MML07] M UELLER C., M ARTIN B., L UMSDAINE A.: A com-
2009, ch. 37. 9 parison of vertex ordering algorithms for large graph visualiza-
tion. Visualization, 2007. APVIS ’07. 2007 6th International
[KSS06] K LUKAS C., S CHREIBER F., S CHWÖBBERMEYER H.: Asia-Pacific Symposium on (5-7 Feb. 2007), 141–148. 9
Coordinated perspectives and enhanced force-directed layout for
the analysis of network motifs. In Proceedings of Asia-Pacific [MMO05] M C P HERSON J., M A K.-L., O GAWA M.: Discover-
Symposium on Information Visualisation (2006), pp. 39–48. 18 ing parametric clusters in social small-world graphs. In Proceed-
ings of the 2005 ACM symposium on Applied computing (2005),
[LBA10] L AMBERT A., B OURQUI R., AUBER D.: Winding pp. 1231–1238. 18
roads: Routing edges into bundles. Computer Graphics Forum
29, 3 (2010), 853–862. 9 [Mun97] M UNZNER T.: H3: laying out large directed graphs in
3D hyperbolic space. In Proceedings of IEEE Symposium on
[LF06] L ESKOVEC J., FALOUTSOS C.: Sampling from large Information Visualization (1997), pp. 2–10, 114. 5
graphs. In Proceedings of 12th ACM SIGKDD Int. Conference
on Knowledge Discovery and data mining (2006), pp. 631–636. [NCA06] N EUMANN P., C ARPENDALE M. S. T., AGARAWALA
4 A.: Phyllotrees: Phyllotactic patterns for tree layout. In Proceed-
ings of the Joint Eurographics and IEEE TCVG Symposium on
[LK07] L ARAMEE R. S., KOSARA R.: Challenges and unsolved Visualization (2006), Eurographics, pp. 59–66. 5, 6
problems. Lecture Notes in Computer Science 4417 (2007), 231–
254. 21 [Nor96] N ORTH S. C.: Incremental layout in DynaDAG. In Pro-
ceedings of the Symposium on Graph Drawing (London, UK,
[LPS∗ 06] L EE B., P LAISANT C., S IMS PARR C., F EKETE J.-D.,
1996), GD ’95, Springer-Verlag, pp. 409–418. 12
H ENRY N.: Task taxonomy for graph visualization. In Proceed-
ings of AVI workshop on BEyond time and errors (2006), pp. 1–5. [OCNF09] OTJACQUES B., C ORNIL M., N OIRHOMME M.,
1, 13 F ELTZ F.: CGD — a new algorithm to optimize space occu-
pation in ellimaps. In Proceedings of the 12th IFIP TC 13 In-
[LRCP07] L EE B., ROBERTSON G. G., C ZERWINSKI M., PARR
ternational Conference on Human-Computer Interaction (2009),
C. S.: CandidTree: visualizing structural uncertainty in similar
pp. 805–818. 6
hierarchies. Information Visualization 6 (December 2007), 233–
246. 20 [OPPROG09] OVALLE -P ERANDONES M. A., P ERIANES -
RODRIGUEZ A., O LMEDA -G OMEZ C.: Hubs and authorities in
[LRP95] L AMPING J., R AO R., P IROLLI P.: A focus+context
a spanish co-authorship network. In Proceedings of International
technique based on hyperbolic geometry for visualizing large hi-
Conference Information Visualisation (2009), pp. 514–518. 17
erarchies. In CHI ’95: Proceedings of the SIGCHI conference
on Human factors in computing systems (New York, NY, USA, [PD08] P OHL M., D IEHL S.: What dynamic network metrics can
1995), ACM Press/Addison-Wesley Publishing Co., pp. 401– tell us about developer roles. In Proceedings of the international
408. 5 workshop on Cooperative and human aspects of software engi-
neering (New York, NY, USA, 2008), ACM, pp. 81–84. 18
[MCH∗ 09] M OSCOVICH T., C HEVALIER F., H ENRY N.,
P IETRIGA E., F EKETE J.-D.: Topology-aware navigation in [PGB02] P LAISANT C., G ROSJEAN J., B EDERSON B. B.:
large networks. In Proceedings of international conference on Spacetree: Supporting exploration in large node link tree, design
Human factors in computing systems (2009), pp. 2319–2328. 14 evolution and empirical evaluation. Information Visualization,
[Mel06] M ELANCON G.: Just how dense are dense graphs in the 2002. INFOVIS 2002. IEEE Symposium on 0 (2002), 57. 16
real world?: a methodological note. In Proceedings of AVI work- [PGS09] P LAISANT C., G RINSTEIN G., S CHOLTZ J.: Visual-
shop on BEyond time and errors: novel evaluation methods for analytics evaluation. IEEE Computer Graphics and Applications
information visualization (New York, NY, USA, 2006), BELIV 29, 3 (2009), 16–17. 22
’06, ACM, pp. 1–7. 3 [PRB08] P OHL M., R EITZ F., B IRKE P.: As time goes by: inte-
[Mey98] M EYER B.: Competitive learning of network diagram grated visualization and analysis of dynamic networks. In Pro-
layout. In Proceedings of IEEE Symposium on Visual Languages ceedings of the working conference on Advanced visual inter-
(September 1998), pp. 56–63. 8 faces (2008), pp. 372–375. 18

submitted to COMPUTER GRAPHICS Forum (1/2011).

T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs 27

[PS06] P ERER A., S HNEIDERMAN B.: Balancing systematic and Man and Cybernetics, IEEE Transactions on 11, 2 (Feb. 1981),
flexible exploration of social networks. IEEE Transactions on Vi- 109–125. 8
sualization and Computer Graphics 12 (September 2006), 693– [SZ00] S TASKO J., Z HANG E.: Focus+context display and nav-
700. 17, 18 igation techniques for enhancing radial, space-filling hierarchy
[PS08] P ERER A., S HNEIDERMAN B.: Integrating statistics and visualizations. In Proceedings of the IEEE Symposium on Infor-
visualization: case studies of gaining clarity during exploratory mation Vizualization (2000), p. 57. 6
data analysis. In Proceeding of SIGCHI conference on Human [SZG∗ 96] S CHAFFER D., Z UO Z., G REENBERG S., BARTRAM
factors in computing systems (2008), pp. 265–274. 22 L., D ILL J., D UBS S., ROSEMAN M.: Navigating hierarchically
[PSD09] P OHL M., S CHMITT M., D IEHL S.: Comparing read- clustered networks through fisheye and full-zoom methods. ACM
ability of graph layouts using eyetracking and task-oriented anal- TOCHI 3, 2 (1996), 162–188. 14
ysis. In Proceedings of Computer Graphics International (2009). [TAS09] T OMINSKI C., A BELLO J., S CHUMANN H.: CGV–an
8, 22 interactive graph visualization system. Computers & Graphics
[Pur97] P URCHASE H. C.: Which aesthetic has the greatest ef- 33, 6 (2009), 660 – 678. 21
fect on human understanding? In Proceedings of International [TAvHS06] T OMINSKI C., A BELLO J., VAN H AM F., S CHU -
Symposium on Graph Drawing (1997), pp. 248–261. 5 MANN H.: Fisheye tree views and lenses for graph visualiza-
[PWL97] PANG A., W ITTENBRINK C., L ODHA . S.: Approaches tion. In Proceedings of IEEE Information Visualization (2006),
to uncertainty visualization. The Visual Computer 13, 8 (Nov. pp. 17–24. 14
1997), 370–390. 20 [TC05] T HOMAS J. J., C OOK K. A. (Eds.): Illuminating the
[RMC91] ROBERTSON G. G., M ACKINLAY J. D., C ARD S. K.: Path: The Research and Development Agenda for Visual Ana-
Cone Trees: animated 3D visualizations of hierarchical informa- lytics. National Visualization and Analytics Center, 2005. 2, 21
tion. In Proceedings of the SIGCHI conference on Human factors [TC06] T HOMAS J., C OOK K.: A visual analytics agenda. Com-
in computing systems (1991), pp. 189–194. 5 puter Graphics and Applications, IEEE 26, 1 (Jan.-Feb. 2006),
[RPD09] R EITZ F., P OHL M., D IEHL S.: Focused animation of 10– 13. 21
dynamic compound graphs. In Proceedings of International Con- [TE10] T ELEA A., E RSOY O.: Image-based edge bundles: Sim-
ference Information Visualisation (2009), pp. 679–684. 12 plified visualization of large graphs. Computer Graphics Forum
[SA06] S HNEIDERMAN B., A RIS A.: Network visualization by 29, 3 (2010), 843–852. 9
semantic substrates. Visualization and Computer Graphics, IEEE [THM∗ 05] T HOMSON J., H ETZLER B., M AC E ACHREN A., G A -
Transactions on 12, 5 (Sept.-Oct. 2006), 733–740. 8 HEGAN M., PAVEL M.: A typology for visualizing uncertainty.
[SB92] S ARKAR M., B ROWN M. H.: Graphical fisheye views In Proceedings of SPIE Conference on Visualization and Data
of graphs. In Proceedings of the SIGCHI conference on Human Analysis 2005 (2005), vol. 5669, pp. 146–157. 20
factors in computing systems (1992), pp. 83 – 91. 14 [TK08] T EKUŠOVÁ T., KOHLHAMMER J.: Visual analysis and
[SCGM00] S TASKO J., C ATRAMBONE R., G UZDIAL M., M C - exploration of complex corporate shareholder networks. In Pro-
DONALD K.: An evaluation of space-filling information visual- ceedings of the IS&T/SPIE Visualization and Data Analysis (jan
izations for depicting hierarchical structures. International Jour- 2008), vol. 6809, p. 68090F. 9, 17
nal of Human-Computer Studies 53, 5 (2000), 663–694. 5 [TS99] T OYODA M., S HIBAYAMA E.: Hyper mochi sheet: a pre-
[Sch08] S CHWÖBBERMEYER H.: Analysis of Biological Net- dictive focusing interface for navigating and editing nested net-
works. Wiley Series on Bioinformatics, Computational Tech- works through a multi-focus distortion-oriented view. In Pro-
niques and Engineering. Wiley, 2008, ch. 5, pp. 85 – 112. 18 ceedings of the SIGCHI conference on Human factors in com-
puting systems (1999), pp. 504–511. 14
[SFL10] S UD A., F ISHER D., L EE H.-P.: Fast dynamic voronoi
treemaps. In Proceedings of International Symposium on Voronoi [TS07] T U Y., S HEN H.-W.: Visualizing changes of hierarchical
Diagrams in Science and Engineering (June 2010), pp. 85–94. 12 data using treemaps. IEEE Transactions on Visualization and
Computer Graphics 13, 6 (2007), 1286–1293. 6, 12
[Shn92] S HNEIDERMAN B.: Tree visualization with tree-maps:
2-D space-filling approach. ACM Transactions on Graphics 11, [TS08a] T EKUŠOVÁ T., S CHRECK T.: Visualizing time-
1 (1992), 92–99. 6 dependent data in multivariate hierarchic plots - design and evalu-
ation of an economic application. In Proceedings of International
[SKM06] S CHRECK T., K EIM D. A., M ANSMANN S.: Regular Conference on Information Visualisation (2008), pp. 143–150. 6,
treemap layouts for visual analysis of hierarchical data. In Pro- 11, 12
ceedings of Spring Conference on Computer Graphics (2006).
[TS08b] T U Y., S HEN H.-W.: Balloon focus: a seamless multi-
11, 12
focus+context method for treemaps. Visualization and Computer
[SLN05] S ARAIYA P., L EE P., N ORTH C.: Visualization of Graphics, IEEE Transactions on 14, 6 (Nov.-Dec. 2008), 1157–
graphs with associated timeseries data. In Proceedings of IEEE 1164. 14, 15, 21
Symposium on Information Visualization (Oct. 2005), pp. 225–
[vH03] VAN H AM F.: Using multilevel call matrices in large
232. 12
software projects. In Information Visualization, 2003. INFOVIS
[SS05] S CHREIBER F., S CHWÖBBERMEYER H.: MAVisto: a 2003. IEEE Symposium on (Oct. 2003), pp. 227 –232. 9
tool for the exploration of network motifs. Bioinformatics 21,
[vHP09] VAN H AM F., P ERER A.: Search, show context, expand
17 (July 2005), 3572–3574. 18
on demand: Supporting large graph exploration with degree-
[SSH09] S CHULZ H.-J., S CHUMANN H., H ADLAK S.: Point- of-interest. IEEE Transactions on Visualization and Computer
based tree representation - a new approach for large hierarchies. Graphics 15, 6 (2009), 953–960. 16, 21
In Proceedings of IEEE Pacific Visualization Symposium (2009), [vHR08] VAN H AM F., ROGOWITZ B.: Perceptual organization
pp. 81–88. 5, 6 in user-generated graph layouts. Visualization and Computer
[STT81] S UGIYAMA K., TAGAWA S., T ODA M.: Methods for Graphics, IEEE Transactions on 14, 6 (Nov.-Dec. 2008), 1333–
visual understanding of hierarchical system structures. Systems, 1339. 9

submitted to COMPUTER GRAPHICS Forum (1/2011).

28 T. von Landesberger, A. Kuijper, T. Schreck, J. Kohlhammer, J.J. van Wijk, J.-D. Fekete, and D.W. Fellner / Visual Analysis of Large Graphs

[vHSD09] VAN H AM F., S CHULZ H.-J., D IMICCO J. M.: Hon- [ZK08] Z IEMKIEWICZ C., KOSARA R.: The shaping of infor-
eycomb: Visual analysis of large scale social networks. In Pro- mation by visual metaphors. IEEE Transactions on Visualization
ceedings of the 12th IFIP TC 13 International Conference on and Computer Graphics 14, 6 (Nov.-Dec. 2008), 1269–1276. 5,
Human-Computer Interaction: Part II (2009), pp. 429–442. 9 22
[vHvW02] VAN H AM F., VAN W IJK J.: Beamtrees: compact vi- [ZMC05] Z HAO S., M C G UFFIN M., C HIGNELL M.: Elastic hi-
sualization of large hierarchies. In Information Visualization, erarchies: combining treemaps and node-link diagrams. In Pro-
2002. INFOVIS 2002. IEEE Symposium on (2002), pp. 93–100. ceedings of IEEE Symposium on Information Visualization (Oct.
5, 6 2005), pp. 57–64. 5, 7, 15
[vHW08] VAN H AM F., WATTENBERG M.: Centrality based vi-
sualization of small world graphs. Computer Graphics Forum 27,
3 (2008), 975–982. 4
[vLdL03] VAN L IERE R., DE L EEUW W.: Graphsplatting: Visu-
alizing graphs as continuous fields. IEEE Transactions on Visu-
alization and Computer Graphics 9 (2003), 206–212. 7
[vLGRS09] VON L ANDESBERGER T., G ÖRNER M., R EHNER
R., S CHRECK T.: A system for interactive visual analysis of
large graphs using motifs in graph editing and aggregation. In
Proceedings of Vision Modeling Visualization Workshop (2009),
pp. 331–339. 17, 18
[vLGS09] VON L ANDESBERGER T., G ÖRNER M., S CHRECK T.:
Visual analysis of graphs with multiple connected components.
In Proceedings of IEEE Symposium on Visual Analytics Science
and Technology (2009), pp. 155–162. 19, 20
[VMCJ10] V IAU C., M C G UFFIN M. J., C HIRICOTA Y., J U -
RISICA I.: The FlowVizMenu and parallel scatterplot matrix:
Hybrid multidimensional visualizations for network exploration.
IEEE Transactions on Visualization and Computer Graphics 16
(2010), 1100–1108. 17
[vWvdW99] VAN W IJK J., VAN DE W ETERING H.: Cushion
treemaps: visualization of hierarchical information. In Proceed-
ings of IEEE Symposium on Information Visualization (1999),
pp. 73–78. 6
[War00] WARE C.: Information visualization: Perception for De-
sign. Morgan Kaufmann, 2000. 13
[Wat06] WATTENBERG M.: Visual exploration of multivariate
graphs. In Proceedings of the SIGCHI conference on Human
Factors in computing systems (2006), pp. 811–819. 18
[WBS∗ 08] WATSON B., B RINK D., S TALLMAN M., D EVA -
JARAN D., R HYNE T.-M., PATEL H.: Visualizing very large
layered graphs with quilts. Technical Report TR-2008-10, NC
State University, 2008. 10
[WCG03] W ONG N., C ARPENDALE S., G REENBERG S.: Edge-
lens: an interactive method for managing edge congestion in
graphs. In Proceedings of IEEE Symposium on Information Vi-
sualization (2003), pp. 51–58. 14
[WD08] W OOD J., DYKES J.: Spatially ordered treemaps. Vi-
sualization and Computer Graphics, IEEE Transactions on 14, 6
(Nov.-Dec. 2008), 1348–1355. 6
[WMC∗ 09] W ONG P. C., M ACKEY P., C OOK K., ROHRER R.,
F OOTE H., W HITING M.: A multi-level middle-out cross-
zooming approach for large graph analytics. In Proceedings of
IEEE Symposium on Visual Analytics Science and Technology
(Oct. 2009), pp. 147–154. 16
[WS03] W HITE S., S MYTH P.: Algorithms for estimating relative
importance in networks. In KDD ’03: Proceedings of the ninth
ACM SIGKDD international conference on Knowledge discovery
and data mining (New York, NY, USA, 2003), ACM, pp. 266–
275. 17
[YKSJ07] Y I J. S., K ANG Y. A ., S TASKO J., JACKO J.: Toward
a deeper understanding of the role of interaction in information
visualization. IEEE Transactions on Visualization and Computer
Graphics 13, 6 (2007), 1224–1231. 13

submitted to COMPUTER GRAPHICS Forum (1/2011).