See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/304611148

Visualizing Group Structures in Graphs: A Survey

Article in Computer Graphics Forum · June 2016

DOI: 10.1111/cgf.12872

CITATIONS READS

70 1,841

3 authors, including:

Corinna Vehlow Fabian Beck

Universität Stuttgart Otto-Friedrich-Universität Bamberg
20 PUBLICATIONS 815 CITATIONS 118 PUBLICATIONS 2,492 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Fabian Beck on 08 March 2019.

The user has requested enhancement of the downloaded file.

Volume xx (200y), Number z, pp. 1–25

Visualizing Group Structures in Graphs: a Survey

Corinna Vehlow, Fabian Beck, and Daniel Weiskopf

VISUS, University of Stuttgart, Germany

2
2 1 2 3
1 2 3 5 4 6 7 8 9 11 12 10 1
1 1
1 2
3 2 3 3
12 3 12 9 11 12 10
5
10 10 9
4
11
5 11 6 5 11 5 4 6 7 8
4 4 12
7 5
9 9 10
8 4
6 9 6 6
11 7
8
8 12 7 8
7 10

(a) Visual node attributes (b) Juxtaposed (c) Superimposed (d) Embedded

Figure 1: Illustrating examples of the four main categories of visualization techniques to explicitly encode different types of
group structures within graph visualizations. (a) Visual node attributes—here color. (b) Juxtaposed—here using an attached
approach. (c) Superimposed—here using a contour approach. (d) Embedded—here using a hybrid approach.

Abstract

Graph visualizations encode relationships between objects. Abstracting the objects into group structures provides
an overview of the data. Groups can be disjoint or overlapping, and might be organized hierarchically. However,
the underlying graph still needs to be represented for analyzing the data in more depth. This work surveys re-
search in visualizing group structures as part of graph diagrams. A particular focus is the explicit visual encoding
of groups, rather than only using graph layout to indicate groups implicitly. We introduce a taxonomy of visu-
alization techniques structuring the field into four main categories: visual node attributes vary properties of the
node representation to encode the grouping, juxtaposed approaches use two separate visualizations, superimposed
techniques work with two aligned visual layers, and embedded visualizations tightly integrate group and graph
representation. The derived taxonomies for group structure and visualization types are also applied to group visu-
alizations of edges. We survey group-only, group–node, group–link, and group–network tasks that are described
in the literature as use cases of group visualizations. We discuss results from evaluations of existing visualiza-
tion techniques as well as main areas of application. Finally, we report future challenges based on interviews we
conducted with leading researchers of the field.
Categories and Subject Descriptors (according to ACM CCS): H.5.2 [Information Interfaces and Presentation]: User
Interfaces—Graphical user interfaces (GUI)

1. Introduction of the graph structure, to highlight central objects, to show

similar objects, and to reveal outliers. The ability of a vi-
Graphs or networks are used to model relationships between sualization to provide these features largely depends on its
objects of any kind. When analyzing graphs exceeding a cer- efficiency to abstract from individual objects into groups or
tain size, however, we do not want to or cannot study each clusters of objects. For instance, applying a random arrange-
object and each relationship connecting two objects individ- ment of visual representatives of objects does not show any
ually. We use visualization to give us a meaningful overview

submitted to COMPUTER GRAPHICS Forum (4/2016).

2 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

Figure 2: Number of publications and distribution of paper types from 1991 to 2015 in our literature collection.

of these groups and largely affects the readability of the vi- group structures, evaluation papers that study the visualiza-
sualization, for node-link representations [Pur02] as well as tion techniques, and application papers that use variants of
for adjacency matrix diagrams [MML07]. In addition to such the techniques in practice (Figure 2, yellow, purple, and pink
groups of objects, also the relations among objects are often bars).
classified into different types; these types need to be visual- Although the body of literature is constantly growing,
ized as well to fully understand the depicted graph informa- the design space for explicitly encoding group structures in
tion. graphs has not yet been surveyed in detail. Existing reports
Indicating groups in the graph by placing similar objects of state of the art focus on other aspects of graph visual-
close to each other implicitly shows some group structures. ization or subproblems: Herman et al. [HMM00] describe
However, it reduces the potentially multi-dimensional con- several approaches that use the hierarchical group struc-
cept of object similarity to a two-dimensional (node-link) ture for navigation and abstraction with a focus on the ap-
or one-dimensional (matrix) layout problem: while similar- plication to graphs. The survey by Brockenauer and Cor-
ity implies closeness, closeness does not necessarily imply nelsen [BC01] contains mainly graph layout algorithms to
similarity; or in other words, close objects are perceived as visualize flat or hierarchical disjoint groups in graphs. Von
similar although their close placement might only be an ar- Landesberger et al. [vLKS∗ 11] survey the area of graph and
tifact of the layout algorithm, edge bundling technique, or tree visualization in general but only occasionally describe
dimensionality reduction. Moreover, groups could not just techniques to represent groups in graphs visually. Saket et
be interpreted as disjoint sets of objects, but might be struc- al. [SSK14] introduce a taxonomy of tasks for group-level
tured hierarchically, might overlap, or might be fuzzy. Im- graph visualization for disjoint groups. General techniques
plicit encodings of group structures lack the ability to unam- to visualize sets and group structures are reviewed by Al-
biguously define group structures and to encode more com- sallakh et al. [AMA∗ 14], however, without discussing the
plex concepts of groups. integration of these techniques into graph visualizations.
Beck et al. [BBDW16] survey visualization techniques of
A growing number of visualization approaches have been graphs that change over time, where in some of them the
developed to overcome these limitations of implicit group group structure of the graph is considered. Furthermore,
encodings. These indicate explicitly which group structures there exist several surveys of general layout algorithms for
are contained in the graph. These group structures can be node-link diagrams [BETT98, DPS02, GFV13]. Elmqvist
either automatically identified by clustering or categoriza- and Fekete [EF10] provide an overview of how to use a hier-
tion algorithms, or imported from an external source of in- archical group structure of objects for navigation and aggre-
formation. The means to visualize the structures explicitly gation in information visualization techniques, such as scat-
are versatile (Figure 1): for instance, the group memberships ter plots, parallel coordinates, and node-link diagrams.
can be encoded in visual node attributes (Figure 1(a)), they
In this paper, we review the state of the art in visualiz-
can be shown in a separate view that is dynamically linked
ing vertex and edge group structures in graphs. It extends
to the graph view (Figure 1(b)), the group encoding can be
our previous publication [VBW15], which focused on ver-
overlaid onto the graph structure (Figure 1(c)), or both graph
tex group structures only and did not discuss task types. We
and group structure can be merged into an embedded repre-
first introduce the area by discussing the background of the
sentation (Figure 1(d)). Describing the large design space
visualized data and, in particular, formulate a consistent ter-
of explicit visual encodings of group structures in graphs
minology (Section 2). We define the scope of the survey and
and classifying existing visualization technique are the main
describe the applied methodology to collect and analyze the
scope of this survey article.
literature (Section 3). As a basis for the techniques that ex-
The literature we collected reveals that already 110 visu- plicitly visualize graphs and groups, we give an overview of
alization techniques showing vertex (97) or edge (19) group implicit layout methods (Section 4). Our main contribution
structures in graphs were published, most of them in the is the classification of explicit visualization techniques of
past decade (Figure 2, green bars). These are accompanied vertex groups into a two-layered taxonomy that we derived
by various papers on graph layout algorithms that highlight from the collected literature of vertex groups. In contrast our

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 3

Table 1: Taxonomy of vertex group structures including the differentiated into crisp and fuzzy . In crisp overlap-
respective numbers of technique papers in our literature col- ping groups, each vertex vi fully belongs to one or more sets
lection. Sk . This belonging can be described, in alternative to the set
notation, by a |V | × K matrix F, where each matrix coeffi-
Group Structure Taxonomy Overlap cient fik ∈ {0, 1} describes if vi belongs to the k-th set Sk
Disjoint Overlapping ( fik = 1) or not ( fik = 0). In contrast, in fuzzy overlapping
Structure

Flat 25 23 groups, vertices vi may belong to different sets Sk to differ-

ent extent. Here, fik ∈ [0, 1] describes to what fraction the
Hierarchical 48 1
vertex vi belongs to set Sk .

Structure: The groups within the graph might be un-

previous publication [VBW15], we apply a similar scheme structured, referred to as flat group structures , or struc-
to further classify and discuss techniques for edge groups tured. While arbitrarily complex group structures are pos-
(Section 6). As a second addition, we collected tasks that sible, we only focus on hierarchical group structures
are described in the literature, abstracted them, and classified because other forms are only rarely used in visualizations
them with respect to the group structure and task type they showing group structure in graphs. We define a hierarchical
refer to (Section 7). We discuss evaluations and applications group structure as a family of sets H = {H0 , H1 , . . . , HL },
of the presented techniques (Sections 8 and 9). Based on in- where each Hl ∈ H is a set of other group elements from H
terviews we conducted with experts of the field, we identify or graph vertices vi ∈ V . These groups represent the inner
major research challenges for vertex groups that could guide elements of a hierarchy where H0 forms the root element.
future research (Section 10). Hence, for all Hl ∈ H where l = 1, . . . , L (i.e., all groups
but the root element), there exists exactly one parent group
The collected, tagged bibliography is available online† in Hl 0 ∈ H (l 0 ∈ {0, . . . , L}) with Hl ∈ Hl 0 ; since also each
an interactive literature browser. Throughout the paper, we graph vertex is contained in exactly one group, the same ap-
use small icons as visual cues within the text summariz- plies to all vi ∈ V (∀vi ∈ V ∃! l 0 ∈ {0, . . . , L} : vi ∈ Hl 0 ).
ing and augmenting terms, figures, and references. For good
comparability, all main figures illustrating the discussed vi- To build a taxonomy of group structures, we consider
sualization techniques show the same data set (i.e., the same overlap and structure as orthogonal concepts. Hence, as
graph and the same groups for each type of group structure). listed in Table 1, both can be combined into four categories:
disjoint flat , overlapping flat , disjoint hi-
erarchical , and overlapping hierarchical .
2. Vertex Group Structures in Graphs For the flat approaches , the group structure is modeled
Group structures occur in different applications of graphs by the family of sets S, whereas the hierarchical taxonomy
structuring the graph vertices in the form of sets, categories, categories require a hierarchical group structure H. In
or hierarchies. In the following, we define vertex group case of disjoint hierarchical groups , the hierar-
structures in graphs including a taxonomy for the types of chical structure H replaces S because the group elements
vertex group structures. We further discuss origins that the of the hierarchy also provide an overlap-free grouping on
groups can arise from. every level of the hierarchy. For overlapping hierarchical
groups , in contrast, both S and H are required to
encode both the overlap of groups and the hierarchy. The
2.1. Definitions numbers in Table 1 show that all categories, except for over-
We first introduce a static graph G = (V, E), which consists lapping hierarchical groups , are covered by various
of a set of vertices V and a set of edges E ⊆ V × V . Vertex visualization techniques, as further discussed below in Sec-
groups within graphs, in general, can be defined as a fam- tion 5.
ily of sets of vertices S = {S1 , . . . , SK }, where each Sk ⊆ V
Graphs can be extended in different directions, for in-
and K denotes the number of groups. Groups can be differ-
stance, to encode directed or weighted edges, to allow multi-
entiated in several ways (Table 1): they can be disjoint or
ple edges between a pair of vertices (multi-graph), or to em-
overlapping, unstructured (flat) or structured (usually, hier-
bed additional multivariate attributes for vertices and edges.
archically).
Graphs may also change over time regarding their topology
Overlap: In disjoint group structures , for all pairs and attributes. For a dynamic graph , the group structure
(Sk1 , Sk2 ), with k1 6= k2 : Sk1 ∩ Sk2 = ∅. Overlapping group can be defined globally over all points in time, i.e., as a
structures , in contrast, contain at least two sets Sk1 and static group structure. Alternatively, a group structure can
Sk2 with Sk1 ∩ Sk2 6= ∅. Overlapping groups can be further be derived for each point in time, i.e., the groups are dy-
namic as well. In the following, techniques that represent a
dynamic graph are additionally marked with ; only when
† http://go.visus.uni-stuttgart.de/groups-in-graphs the group structure changes over time together with its un-

submitted to COMPUTER GRAPHICS Forum (4/2016).

4 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

derlying graph, the technique is marked with . Since there many layout techniques for node-link representations and
is a multitude of possible extensions like dynamic graphs vertex sorting algorithms for matrices that can be used to
that are often orthogonal to the encoding of group structures, implicitly encode the group structure in the node positions.
we do not explicitly reflect them in our definitions. Hadlak Such implicit encoding techniques are briefly summarized
et al. [HSS15] formalize these as multi-faceted graphs and in Section 4 for vertex groups and Section 6 for edge groups
give a general overview of visualization techniques. but are not part of our taxonomy unless the implicit encod-
ing was combined with an explicit encoding. Our taxonomy,
2.2. Origin of Group Structures therefore, comprises only publications that use an explicit
encoding of the group structure.
Graph and group structures only need to be visualized to-
gether when there is a relationship between them, which is
either known beforehand or should be retrieved through the 3.2. Data Collection and Analysis
visual analysis. Group structures can be based on the graph To collect relevant publications for this survey, we first
topology or additional vertex attributes. Without further at- started with a selection of publications that we knew from
tributes required, topology-based group structures are com- previous research and manually inspected the title of all pub-
monly extracted using graph clustering methods [For10]. lications of various information visualization journals and
Such methods try to detect groups of vertices, the so- proceedings:
called community structure or clustering, with a high den-
sity of edges within the groups but low density of edges • Journals
between groups. These methods usually result in disjoint – Computer Graphics Forum
– IEEE Transactions on Visualization and Computer Graphics
flat or hierarchical group structures / , and
– Information Visualization
for some specialized algorithms, crisp or fuzzy overlapping – Journal of Graph Algorithms and Applications
groups / .
• Conferences
When there are other attributes available to describe the – IEEE Pacific Visualization Symposium (PacificVis) [2001–
graph vertices in a specific application, these can be used as 2004: InVis.au; 2005–2007: APVIS]
well to derive a group structure. A categorical attribute di- – IEEE Symposium on Information Visualization (InfoVis)
rectly translates into disjoint flat groups ; but also [since 2006 a special issue of IEEE Transactions on Visu-
overlapping and hierarchical structures might al- alization and Computer Graphics]
ready be encoded explicitly in a set of attributes. If multiple – International Conference on Information Visualisation (IV)
attributes—in particular, numeric ones—should be aggre- – Joint Eurographics–IEEE VGTC Symposium on Visualization
gated, vertices can be grouped based on these multivariate (EuroVis) [1999–2004: VisSym; since 2008 a special issue of
Computer Graphics Forum]
attributes using standard feature-based clustering and classi-
– Symposium on Graph Drawing (GD)
fication algorithms [XW05].
We also looked at the publications cited by relevant papers
3. Scope and Methodology and work that cited these relevant publications. This way, we
could extent our database step by step to retrieve a compre-
To derive a taxonomy of group structure visualizations, we hensive list of publications relevant to our scope, not just
first defined the scope of the survey, collected relevant pub- limited to the above journals and conferences.
lications, and tagged all of them with respect to certain cate-
gories to structure them. This section describes the method- This literature was structured using tagging as a main in-
ology we applied and gives an overview of the collected lit- strument, starting with a list of freely assigned reasonable
erature dataset. tags that are iteratively merged, extended, and grouped to
categories while working through the literature. For further
details, we refer to the survey by Beck et al. [BBDW16],
3.1. Scope whose tagging process we followed.
The scope of our survey is the visualization of group struc-
tures within graphs following the definitions in Section 2.1.
3.3. Literature Dataset
Compared to the previous version of our survey [VBW15],
in this survey we consider not only groups of vertices but To analyze the data, we tagged all publications on vertex
groups of edges as well. Edge groups will be formally intro- groups with respect to several categories starting with the
duced in Section 6. We thereby consider only techniques that paper type. First, we differentiate papers that use only im-
support the visualization of both the group structure and the plicit encoding (tag: layout_technique; 51 papers) from pa-
graph topology. Techniques that visualize only the groups pers that use explicit encoding. For the latter, we distinguish
but not the graph, or vice versa, only the graph were con- application (41), evaluation (7), and technique (97) papers.
sidered out of scope. We further differentiate between im- Moreover, each of the publications is assigned at least one
plicit and explicit visualization of group structures. There are tag for each of the following categories: graph visualization,

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 5

Table 2: Categories and contained tags with descriptions as 4. Implicit Encodings of Vertex Groups
well as the number of technique, evaluation, and application The most common visual representations of graphs are node-
papers using an explicit visualization of vertex group struc- link diagrams (i.e., visual nodes connected by graphical
tures. All icons used in this survey are added to the respective links represent vertices and edges) and adjacency matrices
tags, except for the icon representing coloring approaches , (i.e., rows and columns represent vertices; cells are marked
not listed in the table. if the two respective vertices are connected by an edge). For
tag (category) #T #E #A description
both techniques, the visual representatives of vertices need
97 7 41 total numbers to be positioned on the canvas, i.e., laid out or ordered. By
graph visualization graph visualization paradigm placing related or similar vertices next to each other, group
node-link 86 7 39 node-link representation of the graph structures can be already indicated implicitly. Please note
matrix 10 1 1 matrix representation of the graph
generic 4 1 being applicable to all graph representations
that our taxonomy and the scope of the paper does not cover
group overlap overlap of group these implicit encodings based on vertex positioning. We
disjoint 73 6 28 no overlap only give a brief overview of implicit approaches in this sec-
crisp overlapping 23 1 15 vertices may belong to different groups tion because they are often combined with explicit encod-
fuzzy overlapping 1 vertices may belong to different groups with
different extent ings of groups and part of some of the discussed visualiza-
group structure structure type of group tion techniques. We differentiate between one-dimensional
flat 48 5 26 unstructured and multi-dimensional layout strategies.
hierarchical 49 2 15 groups are hierarchically structured
group visualization visual representation of groups
visual node attribute 11 3 7 properties of node representation vary
juxtaposed 30 1 3 groups and graph visualized separately 4.1. One-Dimensional Layout
superimposed 35 6 23 use of two aligned visual layers
embedded 22 1 integrate group and graph representation One-dimensional layouts are mainly used for adjacency ma-
graph graph properties trix representations to position vertices along one axis.
bipartite 2 1 bipartite or semi-bipartite graph
directed 21 1 5 relations are directed
Groups of vertices that are well connected appear as vi-
dynamic 17 6 graph changes over time sual block structures, given that the vertices are ordered
dynamic_groups 7 group structures changes over time appropriately at the matrix axes [Lii10, MML07]. Often,
generic 45 5 20 none of the other graph attributes applies
multi 1 2 multi-graph
a hierarchical group structure is used to arrange the ver-
multivariate 16 2 8 graph with multivariate attributes tices [EDG∗ 08]; even when using the hierarchical structure,
weighted 11 1 2 edges are weighted
we can still create different sortings by switching the or-
evaluation type of evaluation
algorithmic 8 3 algorithmically using metrics
der of children of a hierarchy element. Some approaches
case study 52 18 application within application domain let users interactively build a subjectively satisfying order
comparison 4 1 1 comparison with other visualization technique
user feedback 7 1 1 collection of user feedback
of rows and columns [Ber11, PDF14] while others solve the
user study 12 7 2 conducting a study involving users sorting problem algorithmically [HF06, MML07]. But also
application application domain for node-link diagrams , one-dimensional layouts are used,
biology 16 20 visualizing biological data
computer 6 1 1 visualizing computer networks
for instance, arranging the nodes on a circle [Hol06] or linear
document 5 3 visualizing documents and text axes [BVB∗ 11].
economy 9 2 visualizing business/ financial/ transport data
media 3 1 visualizing media data
social network 36 12 visualizing social networks (e.g. co-author)
software engineering 22 3 5 visualizing software artefacts 4.2. Multi-Dimensional Layout
sports 3 visualizing sports-related data
In contrast, multi-dimensional layouts are only applica-
ble to node-link diagrams because only node-link di-
agrams allow the free positioning of nodes in a two- or
three-dimensional space. Force-directed layout algorithms,
such as the Fruchterman-Reingold method [FR91] or the
group overlap, group structure, graph type, evaluation, and Kamada-Kawai method [KK89], can reveal groups because
application. Table 2 gives an overview of these tags and the connected nodes are positioned close to each other. Force-
number of technique (#T), evaluation (#E), and application based approaches have been extended in various ways to fur-
(#A) papers for each tag. In addition, it also acts as a legend ther enforce the implicit grouping of nodes for disjoint flat
for icons used throughout the paper. The main tags for our groups [BC01, DKM06, DM14b, Noa07]. In gen-
visualization taxonomy are the tags for the category group eral graph layout algorithms, a generic approach to consider
visualization and further tags to define the subcategories (not disjoint or overlapping groups / is to use pseudo
part of Table 2). Each explicit visualization paper is assigned (dummy) vertices that represent sets of vertices and are con-
to exactly one of the four main group visualizations. Only nected to all contained vertices [EFN99, EH00, GF11]. For
papers that present or evaluate more than one technique are disjoint groups , another method is based on a divide-
assigned more than one group visualization tag if the pre- and-conquer strategy [ACJM03, AMA07b, EF97, FT04]:
sented techniques are of different type. first, a meta-layout is derived for an aggregated graph with

submitted to COMPUTER GRAPHICS Forum (4/2016).

6 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

collapsed groups; then, the vertices of each group are laid 2 2

out independently. For overlapping groups , some ap-
1 1
proaches apply a sequence of different layout algorithms to 3 3
first generate a rough layout that is refined in later steps by 12 12
other algorithms [BALJ06, BCL∗ 07, LDB11, VRW13]. 10 10
5 11 5 11
4 4
9 9
5. Taxonomy of Vertex Group Structure Visualizations 6 6

8 8
There are various visualization techniques that explicitly en- 7 7

code the vertex group structure within the graph visualiza- (a) Pie-chart (b) Icon
tion. Some of the explicit encodings are based on layouts
already implicitly showing group structures. In total, we col- Figure 3: Visualization of overlapping vertex groups
lected 97 explicit visualization technique papers, which we using glyphs and color . (a) Pie-charts encode the fuzzy
categorized according to a hierarchical taxonomy that con- membership degrees . (b) Icons within the node represen-
sists of two layers (Table 3). In the first layer, the four main tatives encode crisp overlapping group memberships .
categories of our taxonomy (illustrated in Figure 1) are vi-
sual node attributes, juxtaposed visualization, superimposed
visualization, and embedded visualization. They are largely
disjoint; only some superimposed and embedded visualiza- colors [Hea96], color is widely used to convey group in-
tion approaches use visual node attributes as additional ex- formation. Each group Sk ∈ S is assigned a color and the
plicit encoding. The second layer further subdivides the cate- nodes of the graph (Figure 1(a)) or group representatives
gories according to main distinguishing visual features. This (e.g., Figures 5(a), (b.1), and (b.2)) are colored respectively.
section describes all categorized techniques following the hi- In total, 40 techniques use visual node attributes, i.e., color
erarchical taxonomy and illustrates them using conceptual (39) and/or glyphs (7), as primary or secondary explicit en-
sketches. coding of the group memberships; all of them are based on
All techniques were additionally tagged with respect to node-link diagrams to represent the graph . Most of these
the type of group structure (see taxonomy of group struc- approaches combine the group encoding with one of the
tures in Section 2.1) that they visualize. The references are other explicit visualization approaches—juxtaposed, super-
therefore marked with the respective icons: flat or hi- imposed, or embedded visualization. Since these other en-
erarchical , disjoint or overlapping . With re- codings usually dominate the visual appearance, we discuss
spect to the type of overlap, by default crisp overlap can them in later subsections in detail but indicate the additional
be assumed if not indicated otherwise; therefore, only the encoding via visual attributes by an icon .
few fuzzy overlapping groups are marked. Table 3 con- We identified 11 techniques that use only color to visu-
trasts both taxonomies by listing all technique papers classi- alize group membership explicitly [Dek01, DS13, DYL∗ 15,
fied into the respective combination of categories. Few tech- DYLL15, IMMS09, LWC∗ 14, NIST12, SGKS15, SKL∗ 14,
niques combine two explicit visualization approaches or can TLTC05, vHW08]. Nodes that belong to only one group
be used for different types of group structures; each of these are simply colored with respect to that group [DS13,
occurs in several cells of the table. Techniques are thereby DYL∗ 15, DYLL15, SGKS15, SKL∗ 14, vHW08]
marked with 1st (2nd ) if the approach represents the pri- (Figure 1(a)). For flat overlapping groups , nodes
mary (secondary) visualization approach of this technique. are represented using glyphs—“graphical objects designed
In particular, color is often used as secondary explicit vi- to convey multiple data values” [War04]. One approach is to
sual mapping of the group structure. As mentioned before, represent vertices as pie charts [IMMS09,LWC∗ 14,NIST12,
techniques (10) visualizing a dynamic graph with a static ST08] with sections colored with respect to the groups the
group structure will be marked with the symbol . Exclu- vertex belongs to [IMMS09, LWC∗ 14]. For crisp overlap-
sively those techniques that visualize dynamic groups in dy- ping groups , the sections of the pie charts have equal
namic graphs (7) will be marked with the symbol . Finally, size [LWC∗ 14, ST08]. In contrast, for fuzzy overlapping
depending on the underlying graph visualization, each tech- groups , they can have different size to encode the fuzzy
nique is classified as node-link representation , matrix rep- membership degrees fik [VRW13] (Figure 3(a)). Paduano
resentation , or hybrid . and Forbes [PF15] indicate crisp group overlaps by
adding a colored border to a node for each group it belongs
to. Another approach for crisp overlapping groups is
5.1. Visual Node Attributes
to represent vertices using boxes that contain icons (Fig-
The association of a vertex with one or more groups can ure 3(b)), such as cross or check marks, in the particu-
be encoded visually by changing the node representation. lar color for all groups they belong to [TLTC05]. Xu et
Although we can easily distinguish no more than about 7 al. [XDC∗ 13] use glyphs to encode the group overlap as well

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 7

Table 3: Visualization techniques classified by our taxonomy of group visualizations and vertex group structures. References
are marked with 1st (2nd ) if the visualization approach is used as primary (secondary) visual mapping for the type of group
structure. Illustrating images are included only for primary visual mappings.

Vertex Group Structure Taxonomy

Overl. hier.
Disjoint flat Overlapping flat Disjoint hierarchical

Color 1
2

3
12

Section 5.1 4
5
10
11
9
6

Figure 1(a) 7
8
Visual node attributes

1st [Dek01, DS13, DYL∗ 15, DYLL15, 1st [–] 1st [–] 1st [–]
SGKS15, SKL∗ 14, vHW08] 2nd [AHRRC11, BT06, BBT06, 2nd [BD05, BD07, KG06, SBG00, 2nd [VRW13]
2nd [BPF14, CDA∗ 14, EHKP14, ET07, DvKSW12, DEKB∗ 14, HRD10, SLAB15]
GHK10,HGK10,HKV14,MH15,SMM13, IMMS09, LQB12, LWC∗ 14, NIST12,
VBAW15, vdEvW14, WWY∗ 15] PF15, TLTC05, VPF∗ 14, XDC∗ 13]

Glyph 1
2

1
2

1
2

3 3 3
12 12 12

Section 5.1 4
5
10
11 4
5
10
11
4
5
10
11
9 9 9
6 6 6

Figure 3 7
8
7
8
7
8

1st [IMMS09, LWC∗ 14, NIST12, 1st [–]

TLTC05] 2nd [VRW13]
2nd [PF15, ST08, XDC∗ 13]

Separate 1
2
3
1
2

4
3

5
10

9
12

11

1 2 3 5 4 6 7 8 9 11 12 10
5 6
2 2 2
8
4 7 12 12 12

Section 5.2.1 1
2

3
10
12

11
6
7
1
3
10
11
1
3
10
11
1
3
10
11
8 5 9 5 9 5 9
9 4 4 4
5 9
4

Figures 4(a)-(b) 7
6
8
11
12
10 7
6

8
7
6

8
7
6

1st [MH15, SMM13, vdEvW14] 1st [SJUS08, ZXQ15] 1st [AKY05, AvHK06, CLLT15, CC07]
Vertex Group Visualization Taxonomy

Attached
Juxtaposed

1 1 1
2 2 2
1 2 3 5 4 6 7 8 9 11 12 10
1 3 3 3
2 5 5 5
3 4 4 4

Section 5.2.2 5
4
6
6
7
6
7
6
7
7 8 8 8
8 9 9 9
9

Figures 4(c)-(d) 11
12
10
11
12
10
11
12
10
11
12
10

st ∗
1 [AZ13, BBV 12, BD08, BD13,
BFBD10, BHW11, BPD11, BMW15,
BSW13, BVB∗ 11, GF03, GZ11, GBD09,
Hol06, HCvW07, NSC05, PvW06, vH03,
vHSD09, VBSW13]
2nd [RMF12]

Line overlay 1
2

1
2

3 3
12 12

Section 5.3.1 4
5
10
11
4
5
10
11
9 9
6 6

Figure 5(a) 7
8
7
8

1st [AHRRC11, PF15, XDC∗ 13]

Contour overlay 2 2 2 2 2
10
2 2
1
2
1
2
Superimposed

1 1 1 1 1 1 1
3 3 3 3 3 10 3 3 3
5 3
12 12 12 12 10 12 12 12

Section 5.3.2 4
5
10
11 4
5
10
11 4
5
10
11
4
5
10
11
4
5 3
3
11
4
5

11
4
5
10
11
5 10

9
12

11
5
10

9
12

11
4 6
9 9 9 9 9 9 9 4 6
6 6
6 6 6 6 6

Figure 5(b) 7
8 7
8
7
8
7
8
7
8
7
8
7
8 7 8
7 8

1st [BPF14, EHKP14, ET07, GHK10, 1st [BT06, BBT06, BT09b, DvKSW12, 1st [BD05, BD07, DGC∗ 05, DHRMM13,
HGK10, HKV14, NIS15, WWY∗ 15] DEKB∗ 14, LQB12, HRD10, ST08, Hol06, KG06, SBG00, YDG∗ 15]
2nd [VBAW15] VPF∗ 14] 2nd [NSC05]

Partitioning 7
12
1
7 12 1
7
12
1
11
4
4 11 2 4 11

Section 5.3.3 8
9
2
8
9
3
8
9
2

6
6 10 6

Figure 6
10
10 3 5 10 3
5 5 5
3 3

∗
1st
[SKB 14, SA06, ZCCB13] 1st
[LSKS10] 1st
[AFH∗ 10, DWS∗ 14, FWD∗ 03,
Hol06]

Node-link 1
2

3 1
2 1
2

3
1
2

3
12 12
12 3 5 10

12 9 11 10

Section 5.4.1 10 4 6
10 5 11
5 11 4
4 9
5 11 7 8
9 4
9 6
6
6 8 3

8 7

Figure 7(a) 7
3
10
8 5 10 5
7
Embedded

st ∗ st ∗ st st
1 [CDA 14, SMER06, VBAW15] 1 [RHR 10, SZPM10] 1 [ASH14, AMA07a, AMA08, AMA09, 1 [VRW13]
AMA11, DM12, DM14a, HN07b, HN07a,
RPD09, SLAB15, vHvW04]

Hybrid 1
2
3
1 2 3
1
2
3
1 2 3 5 10

1
2
1 2 3

5 3
10
9 11 12 10 9 11 12 10

Section 5.4.2 9
11
12
3 5 4 6 7 8
9
11
9 11 12 10 3

5 4 6 7 8
9
11
12
5 4 6 7 8 3 5
10 12 10
5 5 4
10
4 4 6

Figure 7(b) 6
7
8
6
7
8
7
8

1st [HFM07] 1st [HBF08, MZ11] 1st [RMF12]

submitted to COMPUTER GRAPHICS Forum (4/2016).

8 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

1 1 1
2 2 2
1 2 3 5 4 6 7 8 9 11 12 10 3 3 3

2 2 5 5 5
12 12 4 4 4
1 10 1 10 6 6 6
3 3
11 11 7 7 7
9 9 8 8 8
5 5
4 4
9 9 9

6 6 11 11 11
12 12 12
8 8
7 7 10 10 10

(a) Separate—brushing&linking (b) Separate—visual links (c) Attached—linear (d) Attached—radial

Figure 4: Juxtaposed visualization of disjoint hierarchical vertex groups . (a) Brushing and linking and (b) visual links
are used to highlight associated elements of a subhierarchy. In (c)–(d), the hierarchical group structure visualization is aligned
with the graph visualization to connect leaves of the hierarchy to the respective nodes of the graph.

as other metrics by combining different visual channels in- groups defined interactively based on multivariate attributes;
cluding intensity of color, hue, size, and shape. the number of aggregated edges between groups is mapped
Some techniques optimize the color assignment to max- to the width of the link connecting the group nodes. The
imize either the color differences between neighboring techniques by Sallaberry et al. [SMM13] and Meidiana
groups [GHK10, HGK10, LQB12] or the color sta- and Hong [MH15] visualize the evolution of groups
bility between similar groups [HKV14] . Sansen et using a time-line approach and the graph of a selected time
al. [SLAB15] assign similar colors to nested step separate from the group visualization.
groups. Vehlow et al. [VBAW15] developed a Not only the graph but also disjoint hierarchical group
technique for dynamic graphs with dynamic groups. Here, structures are visualized using tree visual-
each dynamic group—rather than each individual group—is ization methods such as axis-parallel [AvHK06] (Fig-
assigned a color to highlight the evolution of groups, where ure 4(b)) or radial [CC07] node-link diagrams, layered ici-
the optimization approach assigns similar hues to similar dy- cle plots [CLLT15], or a treemap [AKY05] (Figure 4(a)).
namic groups. Abello et al. [AKY05, AvHK06] and Cao et al. [CLLT15]
link the group structure view with the graph view via brush-
5.2. Juxtaposed Visualization ing and linking. By selecting a subtree in the hierarchi-
In juxtaposed visualization approaches, the graph G and the cal structure Hl , the user can navigate through the graph
group structure S are visualized next to each other (Fig- as only the respective subgraph will be visualized. ASK-
ure 4). We distinguish between separate juxtaposition, where GraphView [AvHK06] additionally supports an overview of
both visualization layouts are independent from each other, the complete graph using a matrix representation in a
and attached juxtaposition, where the layouts are aligned, third view. VisLink [CC07] arranges two planes showing
e.g., using the same vertex order. We found 30 technique the group structure and the graph in 3D space. Visual links
papers for that category (as primary approach), of which connect internal nodes of the hierarchy, i.e., group nodes
all but 5 visualize disjoint hierarchically structured groups Hl , with all its vertices vi ∈ Hl , respectively. Again, the
( ; compare to Table 3). highlighting—here using visual links—is done only on de-
mand via selection, and hence, only for a selected subtree of
5.2.1. Separate the hierarchy (Figure 4(b)). Schulz et al. [SJUS08]
also make use of visual links between the groups and the
In separate juxtaposed visualizations, the group structure S graph vertices. They visualize semi-bipartite graphs, i.e., bi-
or H is visualized independently of the graph in different partite graphs with possible edges within the bipartite sets
views. Although drawn separately, the juxtaposed visualiza- of vertices. In their visualization of semi-bipartite graphs,
tions are usually linked by interactions (Figure 4(a)) or vi- both vertices and groups are arranged separately on two ver-
sual indicators (Figure 4(b)). In total, we identified 9 sepa- tical axes and linked visually by straight links, where arcs are
rate juxtaposed visualizations—all but one [SJUS08] for dis- used to visualize relations between the vertices. The sorting
joint group structures ; 5 for flat and 4 for hierarchi- of either one of the two axes can be adapted to reduce edge
cal group structures. All identified separate juxtaposed crossings. Zhou et al. [ZXQ15] show the graph and
visualizations show the graph as node-link diagram . the group structure using separate linked views. Within the
Disjoint flat group structures can be visualized radial group structure view, vertices are represented by arcs.
using node-link diagrams [vdEvW14] . Nodes represent Groups are encoded by circles inside the ring region.

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 9

5.2.2. Attached is rotated by 90 degrees). A timeline attached to the left of

each stacked link shows the evolution of the edge weight
In contrast to separate juxtaposed group visualizations, over time.
attached juxtaposed visualizations align the group struc-
ture visualization with the graph visualization. In to- The group structure can also be aligned with the graph vi-
tal, we identified 21 attached juxtaposed visualizations sualization radially, for instance, by positioning the vertices
(as primary approach), all for disjoint hierarchical group along a circle circumference and by surrounding the graph
structures . For the alignment, these approaches visualization with a radial layered icicle plot [GZ11, Hol06,
use the same linear order and place vertices along one HCvW07] (Figure 4(d)), by drawing the hierarchical struc-
axis [AZ13, BPD11, BBV∗ 12, BD13, BVB∗ 11, BHW11, ture as an indented hierarchy in the center of a radial graph
BMW15, BSW13, GF03, GBD09, NSC05, PvW06, PvW08, representation [BHW11] , or by drawing the hierarchical
vH03, vHSD09] (Figures 1(b) and 4(c)) or a circle [BD08, structure on top of the graph visualization [BD08] . The
BFBD10, GZ11, Hol06, HCvW07, VBSW13] (Figure 4(d)). edges of the graph are visualized using arcs [GZ11] or bun-
dled edges [Hol06,HCvW07] (Holten [Hol06] presents three
One approach to visualizing disjoint hierarchical group explicit visualization techniques and, therefore, is referenced
structures is to use a layered icicle plot that is attached to a in three subsections, respectively). Ghou and Zhang [GZ11]
matrix representing the graph [AZ13, BD13, BSW13, GF03, furthermore allow an abstraction of the graph by collapsing
vH03, vHSD09] (Figure 1(b)). The leaves within the ici- inner nodes of the tree. For representing dynamic graphs ,
cle plot have to be aligned with the rows and columns of the graph within the circle needs to be replaced by a se-
the matrix, i.e., the hierarchical structure is used to gen- quence of graphs G := (G1 , . . . , GT ), for instance, arranged
erate a linear ordering of the vertices represented in rows in colored pieces of circle rings in TimeRadarTrees [BD08].
and columns. Most of these techniques support an abstrac- Using TimeSpiderTrees [BFBD10], relations are visually in-
tion of the graph based on the hierarchical structure by dicated by the orientation of shortened links instead of con-
collapsing and expanding groups to aggregate rows and nectedness. In contrast, within the radial layered matrix vi-
columns [AZ13, GF03, vH03, vHSD09]. sualization [VBSW13], edges are represented as color-coded
markers in a polar coordinate system.
Instead of using a matrix as graph representation, the dis-
joint hierarchical group structure can also be aligned with
a node-link representation of the graph (Figure 4(c)). 5.3. Superimposed Visualization
To be aligned with the hierarchy, the nodes need to be ar-
Another method to show the graph and its group structure
ranged linearly; arcs are usually used instead of straight
together is to overlay their representations (Figures 5 and 6).
links to avoid overplotting of nodes and links. The ArcTrees
In this case, the visualizations of the two layers cannot be
approach [NSC05] combines the linear node-link diagram
rendered independently but have to be fully aligned to create
with a one-dimensional treemap: the arcs are attached to the
a meaningful superimposition. We identified 35 technique
leaves of the tree visualization (first of Figure 4(c)). The
papers that superimpose the group structure onto the graph
disjoint hierarchical structure can also be visualized by a
visualization, where 21 of them use color coding as an
node-link diagram [GBD09] (third of Figure 4(c)) or other
additional explicit visual mapping (see also Section 5.1 and
tree visualizations with a linear leaf order [PvW06, PvW08]
Table 3). All of the superimposition techniques are based on
(second of Figure 4(c)). TimeArcTrees [GBD09] extends
two- or three-dimensional node-link diagrams to visualize
these approaches to dynamic graphs . For each time step,
the graph. We differentiate three main categories of overlays:
the vertices are aligned vertically and directed links are
line overlays (3), contour overlays (25), and partitioning ap-
drawn as arcs right (direction is downward) and left (di-
proaches (8).
rection is upward) of the vertices. An aligned tree node-
link diagram attached at the left visualizes the hierarchy.
5.3.1. Line Overlay
Increasing the scalability of the graph representation, other
approaches place the vertices of the graph on two parallel When using lines as an overlay, for each group Sk ∈ S, a
vertical axes and, instead of arcs, straight links between the line of a particular color connects all nodes of that group
two axes visually encode directed graph edges [BBV∗ 12, without interruption [AHRRC11,XDC∗ 13] (Fig-
BPD11, BVB∗ 11]. This technique can be used not only for ure 5(a)). The LineSets approach [AHRRC11] draws a
dynamic graphs [BBV∗ 12, BVB∗ 11] but graph comparison smoothly curved line for each group, where the shortest
as well [BPD11]. To overcome the problem of visual clut- path is computed by an adopted Lin-Kernighan’s traveling
ter for dense graphs, edge bundling [BPD11] or edge splat- salesman heuristic. In contrast, in the approach by Xu et
ting [BVB∗ 11,BBV∗ 12] (i.e., plotting edge density fields) is al. [XDC∗ 13], for each group Sk , all nodes vi ∈ Sk are con-
applied. Another approach that uses straight links instead of nected using a spanning-tree-like shape, which is a general-
arcs stacks the links horizontally either above or below the ization of the LineSets approach. While LineSets can be ap-
hierarchical group structure visualization to indicate their di- plied to any graph layout, the other approach [XDC∗ 13] uses
rection [BMW15] (fourth of Figure 4(c), here the layout multidimensional scaling (MDS) to arrange similar items

submitted to COMPUTER GRAPHICS Forum (4/2016).

10 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

2 2 2 2
10
1
1 1 1
3 3 3 3
12 12 5 12
10
12
10 10 10
5 3 5
5 11 4 11 5 3 11 9 11
4 4
9 9 9 4 6
6
6 6
8
8 7 8 8
7 7 7

(a) Line overlay in combina- (b.1) Disjoint flat (b.2) Overlapping flat (b.3) Disjoint hierarchical
tion with color (b) Contour overlay in combination with or without color

Figure 5: Superimposed visualization of the vertex group structure using (a) line or (b) contour overlays, often in combination
with . For overlapping groups, the contours either overlap or, as in (b.2), nodes are duplicated and connected by visual links.
In contrast, for hierarchical group structures, the contours are nested (b.3).

close to each other, i.e., it combines line overlays with an that are tangent to the area of linked nodes). Vihrovs et
implicit encoding of groups. Within the extended LineSets al. [VPF∗ 14] create contours using a potential field function.
by Paduano and Forbes [PF15] , a line connect- Wu et al. [WWY∗ 15] generate them based on the Voronoi
ing the members of a group replaces the directed links rather treemap using shrinking and smoothing of the Voronoi cells.
than being drawn as curve on top. All nodes connected to a different group are drawn in the
gaps between the contours.
5.3.2. Contour Overlay Contours may be used alone [DGC∗ 05, ST08], in com-
bination with texture [BT09b] (i.e., each group Sk is as-
Groups can also be visualized within node-link dia-
signed a different texture and the contour is filled respec-
grams using closed contours (we identified 25 tech-
tively), or in combination with color coding (all other
nique papers): all nodes vi within the contour are inter-
approaches). When used in combination with color cod-
preted as belonging to the enclosed group Sk ∈ S or
ing, the contour itself can be colored with respect to the
Hl ∈ H (Figure 5(b)). Such contours share the charac-
group it surrounds [DEKB∗ 14] (e.g., Figure 1(c)) or the con-
teristics of set diagrams such as Euler diagrams. Con-
tour is filled with that color [BBT06, BD05, BD07, BPF14,
tour shapes are versatile, for instance, rectangles [DGC∗ 05,
BT06,DvKSW12,EHKP14,ET07,GHK10,HGK10,HKV14,
DHRMM13, HRD10, YDG∗ 15] (Figures 5(b.2) and 5(b.3)),
HRD10, KG06, LQB12, SBG00, VPF∗ 14, WWY∗ 15] (e.g.,
circle sections [ET07] or circles [Hol06,KG06,NIS15] (Fig-
Figures 5(b.1) and 5(b.2)).
ure 5(b.1)), convex hulls [BPF14, ST08, WWY∗ 15], arbi-
trary two-dimensional curves or splines [BBT06, BD05, Contours are so far used to visualize disjoint flat ,
BT06, BT09b, DEKB∗ 14, DvKSW12, EHKP14, GHK10, overlapping flat , and disjoint hierarchical
HGK10, HKV14, LQB12, VPF∗ 14] (Figure 1(c)), or three- group structures (Table 3). For disjoint flat group struc-
dimensional bubbles [BD07, SBG00]. The GMap approach tures [BPF14, EHKP14, ET07, GHK10, HGK10,
[GHK10,HGK10,HKV14] creates a map of contours that are HKV14, NIS15, WWY∗ 15], also the contours are disjoint,
adjacent to each other using a Voronoi tessellation. In con- while the contours representing overlapping group struc-
trast, the contours within MapSets [EHKP14] are generated tures [BT06,BBT06,BT09b,DvKSW12,DEKB∗ 14,
based on non-crossing spanning trees of points belonging to HRD10,LQB12,ST08,VPF∗ 14] intersect. To untangle over-
the same cluster. The trees can be grown to contiguous non- lapping contours, Henry Riche and Dwyer [HRD10] intro-
overlapping regions that are optimized with respect to their duced two techniques for rectangular contour overlays: a
convexity. Other approaches use such spanning trees as well splitting approach (groups with intersections are split up,
but draw the filled contours using texture splatting; a splat drawn as non-overlapping rectangular shapes, and linked by
is defined as radial function for which the transparency in- lines) and a duplication approach (Figure 5(b.2)) (groups
creases with the radius. The eXamine technique [DEKB∗ 14] are represented by overlaid rectangles and nodes contained
uses an extended self-organizing map neuron grid approach in several groups are duplicated and linked visually). For
to lay out nodes and links but also to draw the contours. disjoint hierarchical group structures , the con-
The contours within KelpDiagrams [DvKSW12] are gen- tours or surfaces are nested to encode the hierarchical struc-
erated using a routing algorithm that links elements of ture visually [BD05, BD07, DGC∗ 05, DHRMM13, Hol06,
the same group by constructing minimum cost paths over KG06, SBG00, YDG∗ 15] (e.g., Figure 5(b.3)). The circle
a tangent visibility graph (i.e., a graph including edges contour approach by Holten [Hol06] visualizes edges be-

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 11

overlays, groups are modeled as nodes themselves and are

7 12 1
7 integrated into the graph. In total, we identified 22 technique
12
1
papers, where 18 approaches are based on node-link repre-
11
4
2 4 11
sentations alone (Section 5.4.1) and 4 approaches are hy-
8
9 8 2
brids of node-link and matrix diagrams (Section 5.4.2).
3 9
6
10 6 5.4.1. Node-Link
10
10
5
5 5
3 Using embedded approaches, the groups Sk ∈ S or Hl ∈
3 3
H are drawn as nodes, e.g., using concave shapes (Fig-
(a) Overlapping flat (b) Disjoint hierarchical ure 7(a.1)). Group nodes are connected by visual links if any
of their members are related [CDA∗ 14] , [HN07b,
Figure 6: Superimposed visualization using partitioning of HN07a] . Besides these aggregated edges, only
screen space into (a) vertically aligned or (b) nested regions. edges e ∈ E within each group are visualized in a non-
aggregated way.
While these techniques are static with respect to the
tween groups by links that are bundled based on the hier- group structure visualization, the following approaches sup-
archical structure (we consider edge bundling an implicit port interactive aggregation methods: groups or subtrees
edge grouping technique as discussed in Section 6). Arc- can be collapsed to visualize only the group node but
Trees [NSC05], although classified as juxtaposed attached not the underlying subset of vertices vi ∈ Sk and their
visualization, could be considered a contour approach be- within-group edges (Figures 7(a.2) and 7(a.3)). In Onto-
cause it uses a contour overlay, in particular, rectangles Vis [SMER06] , each node representing a group
nested in one dimension. Sk is connected to all its members vi using visual links in
addition to links encoding the edges of the graph. An ap-
5.3.3. Partitioning proach to visualize the evolution of groups for dynamic
graphs is to draw groups as rectangles on top of a flow-like
Similar to contour overlays, partitioning indicates group
group evolution visualization; between-group edges are ag-
membership by visual enclosing. In contrast to the contour
gregated and the subgraphs of individual groups are drawn
approaches, partitioning is space-filling: the screen space is
within the group representations [VBAW15] .
divided into areas that represent the groups. We identified
Disjoint hierarchical groups can be visualized
8 partitioning approaches, all of them are based on node-
using nested rectangular [ASH14, DM12, DM14a, RPD09]
link diagrams to represent the graph. For disjoint flat
or circular [AMA07a, AMA08, AMA09, AMA11] (3D:
groups , the area of the node-link diagram is par-
spherical [vHvW04]) group structure representations (Fig-
titioned vertically or horizontally into the respective num-
ure 7(a.2)). Reitz et al. [RPD09] use the dynamic hierar-
ber of areas K—one for each group Sk ∈ S [SA06, SKB∗ 14,
chical group structure to control the animation of the dy-
ZCCB13]; nodes are laid out within the area they belong
namic graph visualization and to automatically aggregate
to (Figure 6). Each area is either surrounded by a rectan-
subhierarchies that do not change. In contrast to these tech-
gular contour [SKB∗ 14, ZCCB13] or colored with respect
niques, the Adjasankey diagrams [SLAB15]
to the group it presents [SA06] . If groups overlap, the
use a one-dimensional layout of the nodes aligned twice: all
same approach can be used, but nodes that belong to differ-
nodes that have outgoing links vertically and all nodes with
ent groups are duplicated [LSKS10] (Figure 6(a)).
incoming links horizontally. Edges are drawn as flow-like
Beyond what is shown in the figure, the approach by Lex
rectangular links connecting two nodes. Based on the hierar-
et al. [LSKS10] arranges a two-dimensional area for each
chical structure, nodes and links can be aggregated to group
group in 3D, like walls of a room, and adds visual links be-
nodes and meta edges.
tween shared nodes to visualize the overlap. For disjoint hi-
erarchical structures , the screen is partitioned in a The grid-based visualization approach by Rohrschnei-
space-filling way using a circular icicle plot [AFH∗ 10] or der et al. [RHR∗ 10] arranges the graph nodes on
a treemap approach [DWS∗ 14, FWD∗ 03, Hol06] (see Fig- a regular orthogonal grid, where edges are routed on this
ure 6(b)), where each subsection representing Hl ∈ H is sur- grid using a cost minimization technique. Nodes vi con-
rounded by a contour. tained in different groups Sk are duplicated. In contrast,
Sallaberry et al. [SZPM10] place nodes belong-
ing to at least two groups between the respective group
5.4. Embedded Visualization
nodes, while vertices vi that belong to only one group Sk
The fourth main category of our taxonomy is the embed- can be aggregated and collapsed into group nodes. The ap-
ded visualization of group structures (Figure 7). At a first proach by Vehlow et al. [VRW13] for fuzzy over-
glance, this category looks similar to the superimposition lapping groups is similar: it aggregates vertices vi hierarchi-
approach using contours (Section 5.3.2). But in contrast to cally based on their membership degrees fik (Figure 7(a.3)).

submitted to COMPUTER GRAPHICS Forum (4/2016).

12 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

2 1 2 3 5 10
2
1
2 1 1 2
3 3
1 3
12 12
3 10 5
5
12 9 11 10 10
4 6
10 5 11 9 11 12 10 3
4
9
5 11 7 8 9 3 5 4 6 7 8
4 11
9 6 3 12
6 5 10
8 3

3
7 4
10
8 5 10 5 6
7
7
8

(a.1) Disjoint flat (a.2) Disjoint hierarchical (a.3) Overl. hierarchical (b) Hybrid: node-link &
(a) Node-link-based representation including groups as nodes that can be aggregated matrix

Figure 7: Embedded visualization of vertex groups: (a) Node-link-based integrated representations, where groups are included
as nodes of the graph and can be aggregated. (b) Using a hybrid of a node-link and matrix representation of the graph and
groups within the graph.

Van Ham and Van Wijk [vHvW04] collapse groups by de- edge groups is analogous to that for vertex groups (Sec-
fault and show only the area underneath a lens in more de- tion 3.1), i.e., we only consider techniques that support the
tail. For all other approaches, aggregation is performed by visualization of both the edge group structure and the graph
individually collapsing or expanding group nodes interac- topology. We organized the publications using the same data
tively by clicking on group nodes within the node-link di- collection, analysis, and tagging process as described in Sec-
agram [ASH14, DM12, DM14a, RHR∗ 10, SZPM10] or in a tions 3.2 and 3.3.
separate tree view [AMA07a, AMA08, AMA09, AMA11].
In analogy to vertex group structures, edge groups can
5.4.2. Hybrid: Node-Link and Matrix be defined as a family of sets of edges E = {E1 , . . . , EK },
where each Ek ⊆ E and K denotes the number of groups. We
We identified four approaches that use matrix representa- identified visualizations for disjoint as well as for crisp
tions to visualize edges within groups and links for relations overlapping edge groups that were all unstructured, i.e.,
between groups [HFM07, HBF08, MZ11, RMF12] . In flat . In disjoint flat edge group structures , all pairs
NodeTrix [HFM07] , the adjacency matrices are (Ek1 , Ek2 ), with k1 6= k2 : Ek1 ∩ Ek2 = ∅. Overlapping group
connected to other matrices using edge bundles that visual- structures , in contrast, contain at least two sets Ek1 and
ize the between-group relations. This approach was extended Ek2 with Ek1 ∩ Ek2 6= ∅. In the following, implicit and explicit
to visualize overlapping groups by duplicating vertices vi for visualization techniques for disjoint or overlapping
each group Sk they belong to [HBF08] (Figure 7(b)). flat edge groups will be presented.
Also the approach by Misue and Zhou [MZ11] al-
lows one to visualize overlaps using node duplication. Here, Similar to vertex groups, edge groups can be visualized
in addition to the matrices representing groups, a node is implicitly. Within adjacency matrix representations , edge
drawn for each group and linked to all its members vi , i.e., groups emerge visually as clusters of cells given that the
to the respective rows or columns of the matrices or to sin- vertices are ordered appropriately at the matrix axes [Lii10,
gle nodes vi not contained in any group and hence matrix. MML07]. Using force-directed layout algorithms for node-
TreeMatrix [RMF12] encodes hierarchical struc- link diagrams implicitly shows groups of edges as well.
tures, where subgraphs Hl are shown as adjacency matrices In addition, within node-link diagrams , edge bundling
with an attached hierarchy that is visualized as a node-link methods can be used to deform and visually group similar
diagram or using an icicle plot (see also Section 5.2.2) and edges into bundles. There are three general types of edge
can be collapsed interactively (Figure 1(d)). bundling methods for graphs: cost-based, geometry-based,
and image-based edge bundling methods surveyed by Zhou
6. Edge Group Structure Visualizations et al. [ZXYQ13]. Edges are thereby commonly bundled hi-
erarchically.
The previous part of this survey focused on groups of ver-
tices. Although considering groups based on graph vertices Often, the implicit encoding is combined with an explicit
is more common, also relations can be grouped. This section encoding. Edge bundling is sometimes integrated with the
gives an overview of edge group structures in graphs and visual edge attribute color to explicitly visualize groups of
how these groups and their visualizations fit into the previ- edges. Our taxonomy for edge group visualizations covers
ously defined taxonomies of group structures and group vi- only those edge bundling techniques that use an explicit en-
sualizations. The scope of the survey on visualizations of coding in addition to bundling.

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 13

1 2 3 5 4 6 7 8 9 11 12 10
2 2 2
1
1 1 2 1
3 3 3 3
12 12 5 12
4
10 10 10
6
5 11 5 11 7 5 11
4 4 4
9 9 8 9
9
6 6 6
11
8 8 12 8
7 7 10
7

(a) Visual attribute—shape (b) Visual attribute—color (c) Visual attribute—color (d) Superimposed—line

1 1 1 1 1 1 k 1 2 3 1 2 3 5 4 6 7 8 9 11 12 10
1
2 2 2 2 2 2 2
3 3 3 3 3 3 3

5 5 5 5 5 5
1
4 4 4 4 4 Edge Group 1
4 2
2 12 3
6 6 6 6 6 6
Edge Group 2
10 5
1 3 7 7 7 7 7 7 4
11 Edge Group 3
8 8 8 8 8 8 6

5 9 7
9 9 9 9 9 9
4 8
11 11 11 11 11 11 9
6 11
12 12 12 12 12 12
8 12
7 10 10 10 10 10 10 10

(e) Juxtaposed—attached (f) Juxtaposed—separate (g) Embedded—node-link (h) Embedded—matrix

Figure 8: Visualization techniques for edge groups of disjoint or overlapping flat groups.

There are some visualization techniques that explicitly en- group color if the edge belongs to that group [AZ13, Zec10]
code the edge group structure within the graph visualiza- (Figure 8(c)). As an alternative to color, the style of the
tion. We identified 19 explicit visualization technique pa- links within node-link diagrams can be varied to visualize
pers, which we categorized according to the same hierarchi- which group an edge belongs to [JJ10,Kit03,SGKS15] ,
cal taxonomy as for vertex groups, including the four main e.g., using solid and different types of dashed links (Fig-
categories: visual attributes, juxtapositioning, superposition- ure 8(a)).
ing, and embedding. As mentioned before, only flat edge
We identified three juxtaposed visualization tech-
groups could be identified. Therefore, all techniques pre-
niques [BPD11,DLR10,RMM15], which can all be used for
sented in this section will be tagged with respect to the type
overlapping groups . All three techniques show the graph
of overlap only: disjoint (11 techniques) or overlap-
topology using a node-link diagram . Detangler [RMM15]
ping (8 techniques).
visualizes the representation of groups separately from the
The association of an edge with one or more groups can graph; both views are connected by brushing and link-
be encoded visually by changing the visual edge attributes, ing (Figure 8(e)). In the node-link-based group representa-
i.e., the link in a node-link diagram or cell within a ma- tion, each node represents a group of edges Ek , with re-
trix . The use of visual edge attributes is the most common lations between groups if they share any edges. Other ap-
approach for encoding edge groups, with 13 out of 19 tech- proaches show the graph multiple times—once for each
niques. Among these techniques, 10 make use of color to group of edges [BPD11, DLR10]. Both techniques make
encode disjoint [EHP∗ 11, ETB11, GK07, HF06, TE10, use of a fixed one-dimensional [BPD11] (Figure 8(f)) or
YS15] or overlapping [AZ13, DKL13, YDG∗ 15, Zec10] two-dimensional [DLR10] node-link layout that is reused
groups. In node-link diagrams , links [ETB11] or edge for each group, and hence, graph representation. Didimo et
bundles [EHP∗ 11, GK07, TE10, YS15] are colored with re- al. [DLR10] use color in addition to the juxtaposition.
spect to the group they belong to. If groups overlap , links
We are aware of only one superimposed visualization
that belong to several groups are duplicated and drawn next
technique, in particular, a line overlay approach. Vehlow et
to each other in the respective colors [DKL13, YDG∗ 15]
al. [VHTW13] visualize pairs of edges—i.e., |Ek | = 2—
(Figure 8(b)). Color can also be used for adjacency ma-
within a node-link diagram by connecting the respective
trix representations by coloring the cell with respect
links with curves (Figure 8(d)).
to the group the edge belongs to [HF06]. For overlapping
groups , each cell is subdivided into regions—one for We identified two embedded visualization techniques,
each group—and the region is filled with the respective one that shows the graph as a node-link diagram [PvW08]

submitted to COMPUTER GRAPHICS Forum (4/2016).

14 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

Table 4: Group-related tasks for different types of group structures: disjoint flat, disjoint hierarchical, overlapping flat, and
dynamic group structures. The tasks are grouped with respect to their type: group-only task (GOT), group–vertex task (GVT),
group–edge task (GET), or group–network task (GNT). Tasks that relate to vertex groups or edge groups are marked with VG
or EG, respectively.

Disjoint flat group structures

GOT VG What is the number of groups in S ? [HRD10, JRHT14, SSK14, SSKB15]
EG What is the number of groups in E ? [EHP∗ 11, HF06]
GVT VG Which vertex group Sk is the largest (smallest)? [APP10,DS13,HD12,SSK14] – How large is the size difference between two vertex
groups? [HRD10, JRHT14] – Which vertex group Sk does a given vertex vi belong to? [SSK14, SSKB14, SSKB15] – Which (how
many) vertices are associated with a given group? [AHRRC11, DEKB∗ 14, DvKSW12, PLS∗ 13, SSK14, VKB∗ 15, VPF∗ 14, XDC∗ 13]
([DS13, SSK14, SSKB14]) – Does a set of vertices belong to the same group? [JRHT14, SSK14, SSKB14]
GET EG What is the size of a given edge group Ek ? [EHP∗ 11, HF06] – Which group Ek does a given edge e j belong to? [ETB11] – Which
edges e j are outliers, i.e., isolated edges or missing edges in a group? [HF06]
GNT VG Which one is the vertex with the highest degree in a particular group? [CSL∗ 10, GMT09, HD12, JRHT14, PSK11, SSK14] – Which
vertices of a group are related to vertices of other groups? [GMT09, WWY∗ 15] – Which group is most connected to a particular
vertex or group? [APP10] – Given two vertices A and B, how many groups have to be passed on the path from A to B? [JRHT14,
SSK14] – How many vertices (edges) need to be removed to disconnect two given groups? [SSK14] – Given two vertex groups,
how strongly are they coupled? [CDA∗ 14,SSKB14,WWY∗ 15,YLZ∗ 13] – Given a vertex group, how strongly is this group coupled to
other groups? [CDA∗ 14] – Which vertex group has the maximum number of adjacent groups? [CDA∗ 14,CSL∗ 10,HBF08,LWC∗ 14,
SSK14] – What is the number of edges within a given vertex group? [SSK14] – Which group is the most sparsely (most densely)
connected vertex group? [CSL∗ 10, GMT09, PSK11, SSK14]
EG Which pairs of vertices have a relation from a particular edge group Ek ? [ETB11] – Which vertices are covered by a given edge
group Ek ? [DHRW15] – Which edge groups Ek share many vertices? [DHRW15] – Which edge group Ek is sole vertex connec-
tor? [DHRW15]

Disjoint hierarchical group structures

GOT VG What is the hierarchical structure of the graph? [HN07b] – What are the top-level or bottom-level parts of the hierarchy? [DWS∗ 14]
– Given a group, which subgroups are direct children of this group? [GZ11]
GVT VG Which subgroups Hl is a vertex directly or indirectly allocated to? [GZ11] – Which subgroup is the earliest common parent of a set
vertices? [GZ11] – What are the top-level or bottom-level parts of the hierarchy? [DWS∗ 14]
GNT VG Which subgroups Hl of the hierarchy have a high degree of within-group edges, i.e., which groups are cohesive? [ABZD13,BPD11,
RMF12] – Which subgroups Hl of the hierarchy have a high degree of between-group edges, i.e., which subgroups are cou-
pled? [ABZD13, BPD11, DWS∗ 14] – How are edges of the graph linked to the hierarchical group structure, i.e., does the hierarchi-
cal group structure reflect the graph topology? [BD13, NSC05] – Are there edges between different layers of the hierarchical group
structure? [RMF12] – Which vertices are involved in such cross-layer edges? [RMF12]

Overlapping flat group structures

GVT VG Which groups overlap? [AHRRC11, DEKB∗ 14, LWC∗ 14, VPF∗ 14, XDC∗ 13] – To what extent do groups overlap? [DvKSW12,
HRD10,VPF∗ 14,XDC∗ 13] – Which vertices are associated with only one group? [HBF08,ST08,VRW13] – Which vertices are asso-
ciated with at least two groups? [GMT09,LWC∗ 14,PF15,ST08,SZPM10,VKB∗ 15,VRW13] – Which vertices build bridges between
groups, i.e., vertices whose removal disconnects the groups and makes them disjoint? [BPF14, GMT09, HBF08, LPP∗ 06, VRW13]
– Which groups does a given vertex belong to? [AHRRC11, DEKB∗ 14, DvKSW12, HRD10, VPF∗ 14, VRW13, ZXQ15, XDC∗ 13]
– For fuzzy overlapping groups, this question can be extended to: To what extent does a given vertex contribute to its
groups? [VRW13] – Does a set of particular vertices belong to the same groups? [AHRRC11] – Which two groups share the
largest number of vertices? [HBF08] – Which vertices are in group A and/or B? [DvKSW12, HRD10] – Which vertices are in group
A but not in B? [DvKSW12, HRD10]
GET EG Given multiple groups, what are their common edges? [MGK11] – Given an edge, is it part of multiple groups? [MGK11]
GNT VG How does the overlapping group structure map to the graph topology? [BT06] – Which group is the most central one, i.e., the group
that shares vertices with high degree with the largest number of other groups? [HBF08] – Which groups share the largest number
of vertices with high degree, and hence, have the strongest cohesion? [HBF08]

and one that shows the graph as a matrix representa- (Figure 8(g)). The Dual Adjacency Matrix [DHRW15] visu-
tion [DHRW15]. Pretorius and van Wijk [PvW08] visual- alizes the edge groups integrated into the graph visualization
ize edge groups as nodes positioned in the center of the vi- (Figure 8(h)). The rows (columns) of the top left quadrant
sualization. The edges of the graph are partitioned by letting represent edge groups. The graph is visualized in the bottom
every link pass through the node that represents its group right quadrant of the matrix; rows (columns) can be aggre-

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 15

gated to vertex groups. The two remaining quadrants show Table 4 shows an overview of all collected and abstracted
which vertices (vertex groups) are covered by which edge tasks for each of the four group structure categories. The
groups. tasks are sub-grouped with respect to their type: group-
only task (GOT), group–vertex task (GVT), group–edge task
(GET), or group–network task (GNT). Most of the tasks
relate to vertex groups (marked with VG) and only few
7. Tasks tasks refer to edge groups (marked with EG). In addition,
Depending on the application and the type of group struc- the references from which the tasks were collected are inte-
ture, different tasks are relevant for conducting a visual anal- grated into the table. First, all tasks for disjoint flat
ysis of group structures in a graph. Lee et al. [LPP∗ 06] groups are listed. These also apply to disjoint hierarchi-
present a list of tasks for visualization that are commonly cal groups, considering subgroups Hl rather than
encountered while analyzing graph data. Among these tasks, groups Sk . For overlapping flat and overlapping hi-
they list the identification of clusters (groups) as an im- erarchical groups, these tasks apply as well, where
portant task. Saket et al. [SSK14] introduce a task taxon- some tasks need to be rephrased slightly to consider that a
omy for group-related graph tasks for disjoint flat vertex vertex may belong to different groups. Although not listed
groups . Their tasks concentrate on vertex groups vi- in the table, in analogy, tasks for overlapping hierarchical
sualized within node-link diagrams using a map metaphor— groups are a superset of all tasks listed in the table;
a superimposed contour approach. Following their task tax- so far, no tasks were proposed that require overlapping and
onomy, we group tasks into four categories, including group- hierarchical groups for a single tasks.
only, group–vertex, group–edge, and group–network tasks.
However, we define the task categories slightly different,
8. Evaluation
as described in the following. Group-only tasks can be per-
formed by only considering the groups, i.e., the number K Our collection of publications contains only few evaluation
or L of groups contained in S, E, or H or the nesting struc- papers that describe extensive user studies (8 in total, 6 for
ture of subgroups Hk ∈ H; no vertex or edge information vertex groups and 2 for edge groups), but most of the tech-
is required. For group–vertex tasks, both group and ver- nique papers include some kind of evaluation (see evalua-
tex information has to be considered, i.e., everything de- tion tags in Table 2). In this section, we summarize the re-
scribed by S or H. Group–edge tasks can be performed sults presented in the 8 evaluation papers as well as insights
by only taking group and edge information into account, gained from user studies contained in technique papers that
i.e., everything described by the edge group structure E. For thoroughly evaluate group-related tasks (4 in total).
group–network tasks, all information—group, vertex, and
edge information—has to be considered. This includes the Contrasting visual node attributes (Section 5.1) and su-
graph G and the group structure S, E, or H, respectively. perimposed techniques (Section 5.3), a series of four re-
Therefore, we categorize tasks such as “count the number of cent user studies, by now, provides the most systematic
edges in a given vertex group” as group–network task, rather evaluation of visualization techniques in the field: Saket et
than group–edge task as done by Saket et al. [SSK14], be- al. [SSKB14] compared a superimposed con-
cause these tasks indirectly require the information of which tour approach (GMap [GHK10]) against the use of color
vertices are contained in the groups. as a visual node attribute (Section 5.1). They investigated
several network-based as well as group-based tasks for dis-
The task taxonomy by Saket et al. [SSK14] is limited joint flat group structures including group–vertex tasks and
to disjoint flat vertex group structures . We extend one group–edge task. The results of their user study with
their task taxonomy to cover tasks for disjoint hierarchi- 36 participants suggest that adding contours does not nega-
cal groups and overlapping flat groups . tively impact the performance of network-based tasks and
We also searched for tasks for overlapping hierarchical the GMap approach outperforms colored nodes with re-
groups , but—due to the yet limited coverage of spect to group-based tasks. In a second study, Saket et
such visualizations in literature—we did not find any spe- al. [SSKB15] report on experiments measur-
cific tasks that were not yet covered by the previously men- ing the extent to which people remember the data de-
tioned task categories. To collect tasks for different types picted in these two types of group structure visualiza-
of groups structures, we went through all technique, eval- tions. The 40 participants of the study had to do tasks—
uation, and application papers for vertex or edge groups including one group-only task and one group–vertex task—
in graphs and searched for particular keywords, including four days after being exposed to the visual stimuli. The re-
“task”, “question”, “?”, “identify”, “analyze”, “determine”, sults suggest that participants recall data shown with col-
and “find”. All tasks were abstracted to phrasings using the ored contour-based approach more accurately than using
words vertex, edge, and group, e.g., the task “identify people color only with differences in the accuracy of the tasks
belonging to a particular society” was rephrased to “which performed. Jianu et al. [JRHT14] replicated
vertices are associated with a given group?”. the first study by Saket et al. [SSKB14] and included two

submitted to COMPUTER GRAPHICS Forum (4/2016).

16 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

more approaches in their online study comprising 800 par- approach without duplication. As a result, they found that
ticipants. They evaluated colored nodes, line overlay, and duplications improve group-related tasks but sometimes in-
two types of contour overlays—GMap [GHK10] and Bub- terfere with other graph readability tasks.
bleSets [CPC09]—based on 5 group-based and 5 network-
Hierarchical group structures in graphs have been
based tasks. The group-based tasks included two group-
rarely evaluated, so far; the same applies to juxtaposed ap-
only tasks, one group–vertex task and two group–network
proaches (Section 5.2). There is only one user evaluation on
tasks. With respect to group-based tasks, BubbleSets per-
superimposed contour visualizations in the context of hier-
forms best, followed by lines and the GMap approach, while
archies [FKH15] comparing a hierarchical against
color appears to be least effective.
a flat contour-based visualization. The 29 participants of this
Line overlays (LineSets) were also compared to con- study were asked typical software system comprehension
tour overlays—again the BubbleSets technique—by Alper tasks that do not affiliate in our derived set of group-related
et al. [AHRRC11] . They conducted a user study tasks. With respect to these tasks, they found a statistically
(12 participants) to evaluate the performance on four tasks— significant increase in task correctness of their hierarchical
two group-only tasks, and two group–vertex tasks for over- visualization.
lapping flat groups. Compared to the study by Jianu et
All these evaluations focused on visualizations for ver-
al. [JRHT14], they found that LineSets improve the read-
tex groups. Abuthawabeh et al. [ABZD13] present a study
ability of set membership and set intersection tasks—
evaluating two visualization techniques for overlapping
with higher accuracy rates and shorter completion times—
edge groups: the use of color within the matrix [AZ13,
compared to the BubbleSets technique. They also did an in-
Zec10] (Figure 8(c)) and the juxtaposed attached approach
formal, small-scale eye-tracking study that aimed at under-
as illustrated in Figure 8(f) [BPD11] . Both techniques vi-
standing some of the effects seen in the quantitative results.
sualize the disjoint hierarchical vertex group struc-
Other evaluations focus solely on superimposed contour ture in addition to the edge groups using a juxtaposed at-
approaches (Section 5.3.2): Henry Riche et al. [HRD10] tached visualization of the hierarchy. The tasks of their study
evaluated their Euler diagram technique with respect aim at comparing different edge groups considering the hier-
to its readability considering five tasks—three group-only archical vertex group structure at the same time. They found
tasks and two group–vertex tasks for overlapping flat groups. that all 8 participants were able to identify equivalent edge-
In their study (18 participants), they compared their two rect- groups forming hierarchical groups in the presented graphs.
angular contour overlay techniques—the splitting approach Also Melville et al. [MGK11] compared two color-based
and the node duplication approach (Section 5.3.2)—to a matrix approaches, one that shows all edge groups in one
third (non-convex) contour overlay. They found that the du- matrix (Figure 8(c)) and one that shows the edge groups
plication approach outperforms the other techniques for two juxtaposed using small multiples of the matrix. Based on
of the group-related tasks, but the splitting approach is pre- their study including 18 participants and two group–edge
ferred by many participants. Using a qualitative evaluation, tasks, they found that the comparison of edge groups us-
Byelas and Telea [BT09a] compared algorith- ing one matrix was better than using juxtaposed matrices by
mically generated contour overlays to hand-drawn contours nearly 50%.
to improve the rendering algorithm. The GraphDiaries tech-
nique [BPF14] was evaluated based on a user
9. Application
study comparing it to two other approaches for dynamic
graphs. The focus of the study lies on tasks related to the Group structures occur in various application domains of
dynamic behavior analyzing groups of added or removed el- graphs. In total, comprising application, evaluation, and
ements. technique papers, the most common application domains for
the visualization of vertex group structures are social net-
Some evaluations also take embedded approaches into ac-
works (48 papers), biology (36 papers), and software engi-
count (Section 5.4): Archambault et al. [APP10]
neering (29 papers). For edge groups, biology (9 papers),
compared the use of color as a visual node attribute with
and software engineering (7 papers) can be considered the
an embedded approach, where groups of nodes are replaced
two main application domains. In this section, we summa-
by colored group nodes. They evaluated how this affects the
rize mainly the application papers but occasionally also tech-
readability, but with respect to tasks focusing on attributes
nique papers with a focus on these areas. Further applica-
and graph topology rather than group structures. In contrast,
tion domains of group visualizations are economy networks
Henry et al. [HBF08] evaluated their embedded hy-
representing business, transport, or financial data, computer
brid approach with respect to six tasks—three group–vertex
networks, relations between documents or texts, or relations
tasks, one group–edge task, and two group–network tasks
within media data or sports-related data.
for overlapping flat groups. Their user study (12 partici-
pants) applied different alternatives of vertex duplications In biological applications, graphs are almost exclu-
in overlapping groups and compared these to an embedded sively represented as node-link diagrams . In particular

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 17

in protein-protein-interaction networks and gene correla- ferent types of code couplings, e.g., inheritance, aggrega-
tion networks, disjoint and overlapping flat tion, and usage couplings. These overlapping edge groups
group structures occur. These mainly result from cat- are commonly visualized using the visual edge attribute
egorical attributes of the genes or proteins, e.g., from color [AZ13, ETB11, TE10, Zec10], juxtaposition [BPD11],
cell compartment and pathway associations or from gene or embedding [PvW08].
ontology annotations; also clustering is applied to ex-
tract motifs, i.e., functional groups of proteins. Com-
monly, group structures are visualized by visual node at- 10. Research Challenges
tributes [BST03, DC11, FGB∗ 07, TvDEF09] and superim-
The taxonomy of techniques shows what has been achieved
posed techniques—including overlaid contours [PLS∗ 13,
in the field and reveals possible gaps in the research litera-
RRAS08, SXS∗ 12, VHK∗ 13, VKB∗ 15] and partitioning ap-
ture. However, not necessarily, every gap is a good research
proaches [BMGK08, GFK∗ 14, PK06, SLK∗ 09]. Also at-
opportunity and there might be other interesting challenges
tached juxtaposed [SJUS08] and embedded [RHR∗ 10,
that are not indicated by gaps in the taxonomy. To provide
VRW13] approaches have been applied to biological net-
ideas of worthwhile future research, we discussed research
works. Edge groups in biological networks usually repre-
challenges with respect to vertex group structures with other
sent different types of biological reactions or different con-
researchers who have substantially contributed to the field.
texts where these reactions occur. These disjoint groups
We interviewed 7 experts in graph or group visualization
are commonly visualized using the visual edge attributes
face-to-face—on average for about 40 minutes per person.
color [DC11, LYKB08] or style [GHM∗ 02, JJ10, JKS06,
We first showed a preliminary version of our taxonomy of
Kit03].
vertex group structure visualizations containing illustrations
In social networks such as friendship, communication, of existing techniques (similar to Table 3), explained our in-
collaboration, or co-authorship networks, vertices represent terpretation of vertex groups and group structures in graphs,
people, whereas edges encode relationships between them. and asked them for feedback on terms and definitions—this
Groups of vertices, therefore, identify circles of friends, feedback is already reflected in the terms used in the defini-
groups that cooperate, or the like. Social groups, also tions (Section 2.1) and taxonomy of visualization techniques
called communities, may be disjoint but are often (Section 5). The main purpose of these interviews, however,
modeled more realistically by overlapping groups be- was to ask for the experts’ opinion on open problems and
cause people often participate in a multitude of diverse, challenges on visualizing vertex group structures in graphs.
yet overlapping social communities. So far, social com- Besides challenges they named, we also discussed the chal-
munities have been visualized mainly within node-link di- lenges that we identified beforehand, in case they did not
agrams . Similar to the biological domain, the group mention them already. Based on the feedback we received
structure is commonly visualized by visual node attributes within the interviews and some of our ideas, we identified
(color) [CMF∗ 14, PSK11], superimposed visualization, in five main challenges for vertex groups—each regarded as
particular using line [AHRRC11, XDC∗ 13] or contour over- relevant by 2 to 5 experts. For edge groups, we have not yet
lays [CCC02, DLM14, PS06, SCL∗ 09], or using color and conducted any interviews. In general, there is a remarkably
contour overlays in combination [GMT09, HD12, HB05]. great difference in the number and variety of techniques for
Hierarchical group structures in social networks, in con- edge groups compared to vertex groups. The fact that edge
trast, are often visualized in attached juxtaposed views to- groups have been considered less than vertex groups sug-
gether with static [GZ11] or dynamic graphs [BBV∗ 12, gests that visualizing edge groups is a challenge itself.
GSZ∗ 11]. Also embedded approaches have been applied to
social networks [AMA08, DM14a, HBF08].
10.1. Time-Varying Groups and Comparison
In software engineering, network visualization is used to
analyze program structures and their hierarchical organiza- In many application domains, graphs are not static but
tion , which is usually given by the modularization of change over time, i.e., their topology or their attributes
the software system. Within software architecture diagrams, change over time. It follows that the topology-based or node-
also software metrics can be used to define disjoint attribute-based group structures also change over time. If
or overlapping flat group structures [BT09b, TLTC05]. changes in the graph are significant, the group structure
The hierarchical structure of call graphs or other depen- should be determined for each point in time individually.
dency networks is commonly visualized using attached In contrast, for minor changes in the graph topology or at-
juxtaposed visualizations [BPD11, BD13, PvW07, PvW08, tributes of the graph, it is often sufficient to visualize the
SJSJ05, vH03], superimposed contours [RFG05], or em- static group structure. Most techniques that have been devel-
bedded approaches [PGKG08, RMF12]. Overlapping group oped to visualize the evolution of groups, do not visualize the
structures can be visualized using glyphs [TLTC05] and graph topology [FBS06, OMB∗ 07, RTJ∗ 11, RB10]—for this
overlaid colored [BT09a] or textured [BT09b] contours. reason, these are not part of our taxonomy. Other approaches
Edge groups in software engineering might represent dif- focus on the visualization of dynamic graphs but not the tem-

submitted to COMPUTER GRAPHICS Forum (4/2016).

18 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

poral evolution of groups; they visualize the group structure For instance, superimposed approaches (Section 5.3) be-
aggregated over time. come more and more cluttered through a denser overlay
of group structures. Maybe, even new representations need
First attempts have been made to visualize both the evo-
to be found to handle datasets with large overlap of many
lution of groups and the dynamic graph together either us-
groups. In hierarchical groups , also the depth of the hi-
ing animation [HKV14, KG06, RPD09] or using a timeline-
erarchy could become a problem of scalability for some vi-
based approach [AFH∗ 10, MH15, SMM13, VBAW15], but
sualizations.
these only cover evolving disjoint flat or hierarchical group
structures . Related to dynamic graphs is the prob-
lem of graph comparison: instead of several graphs in a se- 10.4. Interaction Technique
quence, an unordered set of graphs is compared. Similarly,
the comparison of groups structuring these graphs has not One way to address certain issues of scalability is the use of
yet been discussed in this context. interaction methods such as aggregation, which already is a
widely used method—30 of the 111 collected technique pa-
pers support aggregation. But some data is lost through this
10.2. Data Complexity abstraction. The question, therefore, remains how to aggre-
Instead of having multiple versions of the data, the data itself gate while, at the same time, visually encoding the uncer-
can get more complex by adding or refining data dimensions. tainty of aggregated groups and the density of edges within
For overlapping groups , for instance, the visualization these groups. For overlapping groups , aggregation is
of fuzzy memberships is challenging, where vertices even more difficult because overlaps either need to be rep-
may belong to different groups with different extent (see resented explicitly or the overlap is not retrievable for the
Section 2.1). Although the detection of fuzzy overlapping users. Also, there is a need for advanced interactive (semi-)
groups has become quite popular in the domain of graph automatic aggregation methods that guide the user through
clustering [For10], their visualization was only addressed large datasets or define a good default aggregation.
in one work so far [VRW13]. In many applications, groups Beyond aggregation, there is also potential in visual an-
need to express a degree of uncertainty that can be modeled alytics approaches that combine data mining methods with
as fuzzy groups. Another degree of complexity could be in- the visualization of group structures in graphs into an in-
troduced by the topology of the group structure: so far, most teractive approach. Clustering and classification algorithms
of the visualization approaches that were developed for over- could provide alternative group structures on demand. To
lapping groups can handle only flat group structures update the data in a comprehensible way, the visualization
(Table 1). However, also overlapping groups can be orga- needs to adapt on the fly, which introduces new visualization
nized hierarchically , for example, derived from an on- challenges. Similar updates are required when the users edit
tology, through clustering, or other sources. The complexity the group structures interactively, for instance, by applying
of the group structure visualization also increases when mul- set operations to the groups.
tivariate attributes of vertices and edges need to be visualized
together with the graph. These attributes could, for instance,
explain why certain elements are grouped together or why a 10.5. Tasks and Evaluation
pair of groups overlaps.
To choose the right type of group structure visualization for a
particular application, we need to be aware of the tasks users
10.3. Scalability want to solve with the help of the visualization. Application-
specific tasks can be generalized to abstract data tasks, gen-
The data does not need to get more complex, but already vi-
eralizable to different applications. There was already work
sualizing more data elements can be challenging. In graph
done for disjoint flat groups [SSK14], which we extended
visualizations, questions of scalability usually relate to the
to overlapping groups , hierarchical structures , and
number of vertices and density of edges. Visualizing ad-
dynamic groups as well as to edge group structures (Sec-
ditional group structures, however, introduces further chal-
tion 7 summarizes tasks from papers of our bibliography).
lenges. For an increasing number of groups, for instance,
Also, it is important to study how basic data-related tasks
encoding the groups by colors becomes difficult for more
are composed to complex task and which complex tasks are
than about 7 groups [Hea96]. There are already some ap-
most relevant in specific areas of application. Then, it can
proaches that optimize the color assignment (Section 5.1),
be investigated which visualization technique is suitable for
but there is still potential to improve and extend these ap-
which application. Some evaluations have already been con-
proaches. Also, for larger numbers of groups, coloring ap-
ducted (Section 8) but cover the techniques discussed as part
proaches probably need to be replaced by other group repre-
of our taxonomy only partially. Advanced evaluation meth-
sentations.
ods to better understand perceptive and cognitive processes
But even having a constant number of groups, scalability such as such eye tracking [KFBW14] have rarely been ap-
issues could arise from increasing the overlap of groups . plied in the field [JRHT14].

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 19

11. Conclusions [AMA07a] A RCHAMBAULT D., M UNZNER T., AUBER D.:

Grouse: Feature-based, steerable graph hierarchy exploration. In
We presented the state of the art in explicitly visualizing Proceedings of the 9th Joint Eurographics / IEEE VGTC Confer-
vertex or edge group structures in graphs. Groups are dis- ence on Visualization (2007), EuroVis ’07, Eurographics Associ-
joint or overlapping, and might be flat or structured hierar- ation, pp. 67–74. 7, 11, 12
chically. In this survey, we brought together various group [AMA07b] A RCHAMBAULT D., M UNZNER T., AUBER D.:
visualization techniques for graphs that have been discussed TopoLayout: Multilevel graph layout by topological features.
separately, so far. Based on the collected set of publications IEEE Transactions on Visualization and Computer Graphics 13,
2 (2007), 305–317. 5
comprising all these techniques, we derived a taxonomy of
[AMA08] A RCHAMBAULT D., M UNZNER T., AUBER D.:
visualization techniques consisting of four main categories: GrouseFlocks: Steerable exploration of graph hierarchy space.
visual node attributes encode group information in the ap- IEEE Transactions on Visualization and Computer Graphics 14,
pearance of a node, juxtaposed approaches visualize graph 4 (2008), 900–913. 7, 11, 12, 17
and group structure in separate views, superimposed tech- [AMA09] A RCHAMBAULT D., M UNZNER T., AUBER D.: Tug-
niques use visual overlays, and embedded representations Graph: Path-preserving hierarchies for browsing proximity and
combine the graphs and groups into an integrated visualiza- paths in graphs. In Proceedings of the IEEE Pacific Visualization
Symposium (2009), PacificVis ’09, IEEE, pp. 113–120. 7, 11, 12
tion. In addition, we collected and abstracted group-related
tasks described in the papers of our bibliography and cat- [AMA11] A RCHAMBAULT D., M UNZNER T., AUBER D.: Tug-
ging graphs faster: Efficiently modifying path-preserving hierar-
egorized them with respect to the type of group structure chies for browsing paths. IEEE Transactions on Visualization
and type of task including group-only, group–vertex, group– and Computer Graphics 17, 3 (2011), 276–289. 7, 11, 12
edge, and group–network tasks. We intersected the visual- [AMA∗ 14] A LSALLAKH B., M ICALLEF L., A IGNER W.,
ization taxonomy with a taxonomy of group structures to a H AUSER H., M IKSCH S., RODGERS P.: Visualizing sets and
lineup showing which visualization has been already used set-typed data: State-of-the-art and future challenges. In EuroVis
for what type of structure. The comparison hints at oppor- - STARs (2014), EuroVis ’14, Eurographics Association, pp. 1–
21. 2
tunities to fill gaps in existing research literature. Based on
interviews with experts in the field, we identified important [APP10] A RCHAMBAULT D., P URCHASE H. C., P INAUD B.:
The readability of path-preserving clusterings of graphs. Com-
challenges that could guide future research. puter Graphics Forum 29, 3 (2010), 1173–1182. 14, 16
[ASH14] A LTARAWNEH R., S CHULTZ J., H UMAYOUN S. R.:
Acknowledgements CluE: An algorithm for expanding clustered graphs. In Proceed-
ings of the IEEE Pacific Visualization Symposium (2014), Paci-
We would like to thank all experts of our interviews for ficVis ’14, IEEE, pp. 233–237. 7, 11, 12
the feedback on our taxonomy and the discussions of future [AvHK06] A BELLO J., VAN H AM F., K RISHNAN N.: ASK-
challenges: Bilal Alsallakh, Daniel Archambault, Michael GraphView: A large scale graph visualization system. IEEE
Burch, Carsten Görg, Stephen G. Kobourov, Frank van Ham, Transactions on Visualization and Computer Graphics 12, 5
and Jarke J. van Wijk. This work was supported by DFG (2006), 669–676. 7, 8
within SFB 716/D.5. [AZ13] A BUTHAWABEH A., Z ECKZER D.: IMMV: An interac-
tive multi-matrix visualization for program comprehension. In
Proceedings of the 1th IEEE Working Conference on Software
References Visualization (2013), VISSOFT ’13, IEEE, pp. 1–4. 7, 9, 13, 16,
[ABZD13] A BUTHAWABEH A., B ECK F., Z ECKZER D., D IEHL 17
S.: Finding structures in multi-type code couplings with node- [BALJ06] B OURQUI R., AUBER D., L ACROIX V., J OURDAN F.:
link and matrix visualizations. In Proceedings of the 1st IEEE Metabolic network visualization using constraint planar graph
Working Conference on Software Visualization (2013), VISSOFT drawing algorithm. In Proceedings of the 10th International
’13, IEEE, pp. 1–10. 14, 16 Conference on Information Visualization (2006), IV ’06, IEEE,
[ACJM03] AUBER D., C HIRICOTA Y., J OURDAN F., pp. 489–496. 6
M ELANÇON G.: Multiscale visualization of small world [BBDW16] B ECK F., B URCH M., D IEHL S., W EISKOPF D.: A
networks. In Proceedings of the 9th Symposium on Information taxonomy and survey of dynamic graph visualization. Computer
Visualization (2003), IV ’03, IEEE, pp. 75–81. 5 Graphics Forum (2016), to appear. 2, 4
[AFH∗ 10] A HMED A., F U X., H ONG S.-H., N GUYEN Q., X U [BBT06] B YELAS H., B ONDAREV E., T ELEA A.: Visualiza-
K.: Visual analysis of history of world cup: A dynamic network tion of areas of interest in component-based system architec-
with dynamic hierarchy and geographic clustering. In Visual In- tures. In Proceedings of the 32nd EUROMICRO Conference on
formation Communication. Springer, 2010, pp. 25–39. 7, 11, 18 Software Engineering and Advanced Applications (2006), SEAA
[AHRRC11] A LPER B., H ENRY R ICHE N., R AMOS G., C ZER - ’06, IEEE, pp. 160–169. 7, 10
WINSKI M.: Design study of linesets, a novel set visualization [BBV∗ 12] B ECK F., B URCH M., V EHLOW C., D IEHL S.,
technique. IEEE Transactions on Visualization and Computer W EISKOPF D.: Rapid serial visual presentation in dynamic graph
Graphics 17, 12 (2011), 2259–2267. 7, 9, 14, 16, 17 visualization. In Proceedings of the Symposium on Visual Lan-
[AKY05] A BELLO J., KOBOUROV S. G., Y USUFOV R.: Visu- guages and Human-Centric Computing (2012), VL/HCC ’12,
alizing large graphs with compound-fisheye views and treemaps. IEEE, pp. 185–192. 7, 9, 17
In Proceedings of the 12th International Symposium on Graph [BC01] B ROCKENAUER R., C ORNELSEN S.: Drawing clusters
Drawing, vol. 3383 of GD ’04. Springer, 2005, pp. 431–441. 7, and hierarchies. In Drawing Graphs, Methods and Models.
8 Springer, 2001, pp. 193–227. 2, 5

submitted to COMPUTER GRAPHICS Forum (4/2016).

20 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

[BCL∗ 07] B OURQUI R., C OTTRET L., L ACROIX V., AUBER D., [BT09a] B YELAS H., T ELEA A.: Towards realism in drawing
M ARY P., S AGOT M.-F., J OURDAN F.: Metabolic network visu- areas of interest on architecture diagrams. Journal of Visual Lan-
alization eliminating node redundance and preserving metabolic guages & Computing 20, 2 (2009), 110–128. 16, 17
pathways. BMC Systems Biology 1, 1 (2007), 1–19. 6 [BT09b] B YELAS H., T ELEA A.: Visualizing metrics on areas
[BD05] BALZER M., D EUSSEN O.: Exploring relations within of interest in software architecture diagrams. In Proceedings of
software systems using treemap enhanced hierarchical graphs. In the IEEE Pacific Visualization Symposium (2009), PacificVis ’09,
Proceedings of the 3rd IEEE International Workshop on Visualiz- IEEE, pp. 33–40. 7, 10, 17
ing Software for Understanding and Analysis (2005), VISSOFT [BVB∗ 11] B URCH M., V EHLOW C., B ECK F., D IEHL S.,
’05, IEEE, pp. 1–6. 7, 10 W EISKOPF D.: Parallel edge splatting for scalable dynamic
[BD07] BALZER M., D EUSSEN O.: Level-of-detail visualization graph visualization. IEEE Transactions on Visualization and
of clustered graph layouts. In Proceedings of the 6th Interna- Computer Graphics 17, 12 (2011), 2344–2353. 5, 7, 9
tional Asia-Pacific Symposium on Visualization (2007), APVIS [CC07] C OLLINS C., C ARPENDALE S.: VisLink: Revealing re-
’07, IEEE, pp. 133–140. 7, 10 lationships amongst visualizations. IEEE Transactions on Visu-
[BD08] B URCH M., D IEHL S.: TimeRadarTrees: Visualizing dy- alization and Computer Graphics 13, 6 (2007), 1192–1199. 7,
namic compound digraphs. Computer Graphics Forum 27, 3 8
(2008), 823–830. 7, 9 [CCC02] C HANG C.-L., C HEN D.-Y., C HUANG T.-R.: Brows-
[BD13] B ECK F., D IEHL S.: Visual comparison of software ar- ing newsgroups with a social network analyzer. In Proceedings
chitectures. Information Visualization 12, 2 (2013), 178–199. 7, of the 6th International Conference on Information Visualisation
9, 14, 17 (2002), IV ’02, IEEE, pp. 750–755. 17
[Ber11] B ERTIN J.: Semiology of Graphics: Diagrams, Networks, [CDA∗ 14] C HATURVEDI S., D UNNE C., A SHKTORAB Z.,
Maps. ESRI Press, 2011. 5 Z ACHARIAH R., S HNEIDERMAN B.: Group-in-a-box meta-
layouts for topological clusters and attribute-based groups:
[BETT98] BATTISTA G. D., E ADES P., TAMASSIA R., T OL - Space-efficient visualizations of network communities and their
LIS I. G.: Graph Drawing: Algorithms for the Visualization of ties. Computer Graphics Forum 33, 8 (2014), 52–68. 7, 11, 14
Graphs, 1st ed. Prentice Hall PTR, 1998. 2
[CLLT15] C AO N., L IN Y.-R., L I L., T ONG H.: g-Miner: Inter-
[BFBD10] B URCH M., F RITZ M., B ECK F., D IEHL S.: Time- active visual group mining on multivariate graphs. In Proceed-
SpiderTrees: A novel visual metaphor for dynamic compound ings of the 33rd Annual ACM Conference on Human Factors in
graphs. In Proceedings of the Symposium on Visual Languages Computing Systems (2015), CHI ’15, ACM, pp. 279–288. 7, 8
and Human-Centric Computing (2010), VL/HCC ’10, IEEE,
[CMF∗ 14] C RNOVRSANIN T., M UELDER C. W., FARIS R.,
pp. 168–175. 7, 9
F ELMLEE D., M A K.-L.: Visualization techniques for categor-
[BHW11] B URCH M., H ÖFERLIN M., W EISKOPF D.: Lay- ical analysis of social networks with multiple edge sets. Social
ered TimeRadarTrees. In Proceedings of the 15th International Networks 37 (2014), 56–64. 17
Conference on Information Visualisation (2011), IV ’11, IEEE,
[CPC09] C OLLINS C., P ENN G., C ARPENDALE M. S. T.: Bub-
pp. 18–25. 7, 9
ble Sets: Revealing set relations with isocontours over existing
[BMGK08] BARSKY A., M UNZNER T., G ARDY J., K INCAID visualizations. IEEE Transactions on Visualization and Com-
R.: Cerebral: Visualizing multiple experimental conditions on puter Graphics 15, 6 (2009), 1009–1016. 16
a graph with biological context. IEEE Transactions on Visualiza- [CSL∗ 10] C AO N., S UN J., L IN Y.-R., G OTZ D., L IU S., Q U
tion and Computer Graphics 14 (2008), 1253–1260. 17 H.: FacetAtlas: Multifaceted visualization for rich text corpora.
[BMW15] B URCH M., M UNZ T., W EISKOPF D.: Edge- IEEE Transactions on Visualization and Computer Graphics 16,
stacked timelines for visualizing dynamic weighted digraphs. In 6 (2010), 1172–1181. 14
Proceedings of the 6th International Conference on Informa- [DC11] DAVIS D. A., C HAWLA N. V.: Exploring and exploit-
tion Visualization Theory and Applications (2015), IVAPP ’15, ing disease interactions from multi-relational gene and phenotype
SciTePress, pp. 93–100. 7, 9 networks. PLoS ONE 6, 7 (07 2011), e22670. 17
[BPD11] B ECK F., P ETKOV R., D IEHL S.: Visually exploring [Dek01] D EKKER A. H.: Visualisation of social networks using
multi-dimensional code couplings. In Proceedings of the 6th CAVALIER. In Proceedings of the Australasian Symposium on
IEEE International Workshop on Visualizing Software for Un- Information Visualisation (2001), APVIS ’01, IEEE, pp. 49–55.
derstanding and Analysis (2011), VISSOFT ’11, IEEE, pp. 1–8. 6, 7
7, 9, 13, 14, 16, 17
[DEKB∗ 14] D INKLA K., E L -K EBIR M., B UCUR C.-I.,
[BPF14] BACH B., P IETRIGA E., F EKETE J.-D.: GraphDi- S IDERIUS M., S MIT M., W ESTENBERG M. A., K LAU G.:
aries: Animated transitions and temporal navigation for dynamic eXamine: Exploring annotated modules in networks. BMC
networks. IEEE Transactions on Visualization and Computer Bioinformatics 15, 201 (2014), 1–14. 7, 10, 14
Graphics 20, 5 (2014), 740–754. 7, 10, 14, 16
[DGC∗ 05] D OGRUSOZ U., G IRAL E., C ETINTAS A., C IVRIL
[BST03] B REITKREUTZ B.-J. J., S TARK C., T YERS M.: Os- A., D EMIR E.: A compound graph layout algorithm for bio-
prey: a network visualization system. Genome Biology 4, 3 logical pathways. In Proceedings of the 12th International Sym-
(2003), 1–6. 17 posium on Graph Drawing, vol. 3383 of GD ’04. Springer, 2005,
[BSW13] B URCH M., S CHMIDT B., W EISKOPF D.: A matrix- pp. 442–447. 7, 10
based visualization for exploring dynamic compound digraphs. [DHRMM13] DWYER T., H ENRY R ICHE N., M ARRIOTT K.,
In Proceedings of the 17th International Conference on Informa- M EARS C.: Edge compression techniques for visualization of
tion Visualisation (2013), IV ’13, IEEE, pp. 66–73. 7, 9 dense directed graphs. IEEE Transactions on Visualization and
[BT06] B YELAS H., T ELEA A.: Visualization of areas of interest Computer Graphics 19, 12 (2013), 2596–2605. 7, 10
in software architecture diagrams. In Proceedings of the ACM [DHRW15] D INKLA K., H ENRY R ICHE N., W ESTENBERG M.:
Symposium on Software Visualization (2006), SoftVis ’06, ACM, Dual Adjacency Matrix: Exploring link groups in dense net-
pp. 105–114. 7, 10, 14 works. Computer Graphics Forum 34, 3 (2015), 311–320. 14

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 21

[DKL13] DAI B. T., K WEE A., L IM E.-P.: ViStruclizer: A struc- [EF10] E LMQVIST N., F EKETE J.-D.: Hierarchical aggregation
tural visualizer for multi-dimensional social networks. In Ad- for information visualization: Overview, techniques, and design
vances in Knowledge Discovery and Data Mining, Pei J., Tseng guidelines. IEEE Transactions on Visualization and Computer
V. S., Cao L., Motoda H., Xu G., (Eds.), vol. 7819 of Lecture Graphics 16, 3 (2010), 439–454. 2
Notes in Computer Science. Springer, 2013, pp. 49–60. 13
[EFN99] E ADES P., F ENG Q.-W., NAGAMOCHI H.: Drawing
[DKM06] DWYER T., KOREN Y., M ARRIOTT K.: IPSep-CoLa: clustered graphs on an orthogonal grid. Journal of Graph Algo-
An incremental procedure for separation constraint layout of rithms and Applications 3, 4 (1999), 3–29. 5
graphs. IEEE Transactions on Visualization and Computer
Graphics 12, 5 (2006), 821–828. 5 [EH00] E ADES P., H UANG M. L.: Navigating clustered graphs
using force-directed methods. Journal of Graph Algorithms and
[DLM14] D IDIMO W., L IOTTA G., M ONTECCHIANI F.: Net- Applications 4, 3 (2000), 157–181. 5
work visualization for financial crime detection. Journal of Vi-
sual Languages & Computing 25, 4 (2014), 433–451. 17 [EHKP14] E FRAT A., H U Y., KOBOUROV S. G., P UPYREV S.:
MapSets: Visualizing embedded and clustered graphs. In Pro-
[DLR10] D IDIMO W., L IOTTA G., ROMEO S. A.: Graph visual- ceedings of the 22nd International Symposium on Graph Draw-
ization techniques for conceptual web site traffic analysis. In Pro- ing, vol. 8871 of GD ’14. Springer, 2014, pp. 452–463. 7, 10
ceedings of the IEEE Pacific Visualization Symposium (March
2010), PacificVis ’10, pp. 193–200. 13 [EHP∗ 11] E RSOY O., H URTER C., PAULOVICH F. V.,
C ANTAREIRO G., T ELEA A.: Skeleton-based edge bundling
[DM12] D IDIMO W., M ONTECCHIANI F.: Fast layout computa- for graph visualization. IEEE Transactions on Visualization and
tion of hierarchically clustered networks: Algorithmic advances Computer Graphics 17, 12 (2011), 2364–2373. 13, 14
and experimental analysis. In Proceedings of the 16th Interna-
tional Conference on Information Visualisation (2012), IV ’12, [ET07] E LMQVIST N., T SIGAS P.: CiteWiz: a tool for the visual-
IEEE, pp. 18–23. 7, 11, 12 ization of scientific citation networks. Information Visualization
6, 3 (2007), 215–232. 7, 10
[DM14a] D IDIMO W., M ONTECCHIANI F.: Fast layout compu-
tation of clustered networks: Algorithmic advances and experi- [ETB11] E RDEMIR U., T EKIN U., B UZLUCA F.: E-Quality: A
mental analysis. Information Sciences 260 (2014), 185–199. 7, graph based object oriented software quality visualization tool. In
11, 12, 17 Proceedings of the 6th IEEE International Workshop on Visualiz-
ing Software for Understanding and Analysis (2011), VISSOFT
[DM14b] D KAKI T., M OTHE J.: An energy-based model to op- ’11, pp. 1–8. 13, 14, 17
timize cluster visualization. In Proceedings of the 8th Interna-
tional Conference on Research Challenges in Information Sci- [FBS06] FALKOWSKI T., BARTELHEIMER J., S PILIOPOULOU
ence (2014), RCIS ’14, IEEE, pp. 1–11. 5 M.: Mining and visualizing the evolution of subgroups in so-
cial networks. In Proceedings of IEEE/WIC/ACM International
[DPS02] D ÍAZ J., P ETIT J., S ERNA M.: A survey of graph layout Conference on Web Intelligence (2006), WI ’06, IEEE, pp. 52–
problems. ACM Computing Surveys 34, 3 (2002), 313–356. 2 58. 17
[DS13] D UNNE C., S HNEIDERMAN B.: Motif simplification: [FGB∗ 07] F REEMAN T. C., G OLDOVSKY L., B ROSCH M., VAN
Improving network visualization readability with fan, connector, D ONGEN S., M AZIÉRE P., G ROCOCK R. J., F REILICH S.,
and clique glyphs. In Proceedings of the SIGCHI Conference on T HORNTON J., E NRIGHT A. J.: Construction, visualisation, and
Human Factors in Computing Systems (2013), CHI ’13, ACM, clustering of transcription networks from microarray expression
pp. 3247–3256. 6, 7, 14 data. PLoS Computational Biology 3, 10 (2007). 17
[DvKSW12] D INKLA K., VAN K REVELD M. J., S PECKMANN [FKH15] F ITTKAU F., K RAUSE A., H ASSELBRING W.: Hier-
B., W ESTENBERG M. A.: Kelp Diagrams: Point set membership archical software landscape visualization for system comprehen-
visualization. Computer Graphics Forum 31, 3 (2012), 875–884. sion: A controlled experiment. In Proceedings of the 3rd IEEE
7, 10, 14 Working Conference on Software Visualization (2015), VISSOFT
[DWS∗ 14] DANIEL D. T., W UCHNER E., S OKOLOV K., ’15, IEEE, pp. 36–45. 16
S TAL M., L IGGESMEYER P.: Polyptychon: A hierarchically- [For10] F ORTUNATO S.: Community detection in graphs. Physics
constrained classified dependencies visualization. In Proceed- Reports 486, 3-5 (2010), 75–174. 4, 18
ings of the 2nd IEEE Working Conference on Software Visualiza-
tion (2014), VISSOFT ’14, IEEE, pp. 83–86. 7, 11, 14 [FR91] F RUCHTERMAN T. M. J., R EINGOLD E. M.: Graph
drawing by force-directed placement. Software: Practice and Ex-
[DYL∗ 15] D U X., Y E Y., L AU R. Y., L I Y., H UANG X.: Multi- perience 21, 11 (1991), 1129–1164. 5
opinion ring: visualizing and predicting multiple opinion orienta-
tions in online social media. Multimedia Tools and Applications [FT04] F RISHMAN Y., TAL A.: Dynamic drawing of clustered
(2015), 1–28. 6, 7 graphs. In Proceedings of the 10th IEEE Symposium on Informa-
tion Visualization (2004), InfoVis ’04, IEEE, pp. 191–198. 5
[DYLL15] D U X., Y E Y., L AU R. Y., L I Y.: OpinionRings:
Inferring and visualizing the opinion tendency of socially con- [FWD∗ 03] F EKETE J.-D., WANG T. D., DANG N., A RIS A.,
nected users. Decision Support Systems 75 (2015), 11–24. 6, P LAISANT C.: Overlaying graph links on treemaps. In Proceed-
7 ings of the 9th IEEE Symposium on Information Visualization,
Poster Compendium (2003), InfoVis ’03, IEEE. 7, 11
[EDG∗ 08] E LMQVIST N., D O T.-N., G OODELL H., H ENRY N.,
F EKETE J.-D.: ZAME: Interactive large-scale graph visualiza- [GBD09] G REILICH M., B URCH M., D IEHL S.: Visualizing the
tion. In Proceedings of the IEEE Pacific Visualization Symposium evolution of compound digraphs with TimeArcTrees. Computer
(2008), PacificVis ’08, IEEE, pp. 215–222. 5 Graphics Forum 28, 3 (2009), 975–982. 7, 9
[EF97] E ADES P., F ENG Q.-W.: Multilevel visualization of clus- [GF03] G HONIEM M., F EKETE J.-D.: Matrix view of graphs and
tered graphs. In Proceedings of the 4th International Sympo- direct manipulation of cluster hierarchies. In Proceedings of the
sium on Graph Drawing, vol. 1190 of GD ’96. Springer, 1997, 15th French-speaking Conference on Human-computer Interac-
pp. 101–112. 5 tion (2003), IHM ’03, ACM, pp. 206–207. 7, 9

submitted to COMPUTER GRAPHICS Forum (4/2016).

22 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

[GF11] G IBSON H., FAITH J.: Node-attribute graph layout for [HGK10] H U Y., G ANSNER E. R., KOBOUROV S. G.: Visualiz-
small-world networks. In Proceedings of the 15th International ing graphs and clusters as maps. IEEE Computer Graphics and
Conference on Information Visualisation (2011), IV ’11, IEEE, Applications 30, 6 (2010), 54–66. 7, 8, 10
pp. 482–487. 5 [HKV14] H U Y., KOBOUROV S. G., V EERAMONI S.: Embed-
[GFK∗ 14] G ERASCH A., FABER D., K ÜNTZER J., N IERMANN ding, clustering and coloring for dynamic maps. Journal of
P., KOHLBACHER O., L ENHOF H.-P., K AUFMANN M.: BiNA: Graph Algorithms and Applications 18, 1 (2014), 77–109. 7,
a visual analytics tool for biological network data. PLoS ONE 9, 8, 10, 18
2 (2014), e87397. 17 [HMM00] H ERMAN I., M ELANCON G., M ARSHALL M.: Graph
[GFV13] G IBSON H., FAITH J., V ICKERS P.: A survey of two- visualization and navigation in information visualization: A sur-
dimensional graph layout techniques for information visualisa- vey. IEEE Transactions on Visualization and Computer Graphics
tion. Information Visualization 12, 3-4 (2013), 324–357. 2 6, 1 (2000), 24–43. 2
[GHK10] G ANSNER E. R., H U Y., KOBOUROV S. G.: GMap: [HN07a] H UANG M. L., N GUYEN Q. V.: Navigating large clus-
Visualizing graphs and clusters as maps. In Proceedings of the tered graphs with triple-layer display. In Proceedings of the 11th
IEEE Pacific Visualization Symposium (2010), PacificVis ’10, International Conference on Information Visualization (2007),
IEEE, pp. 201–208. 7, 8, 10, 15, 16 IV ’07, IEEE, pp. 684–692. 7, 11
[GHM∗ 02] G OESMANN A., H AUBROCK M., M EYER F., K ALI - [HN07b] H UANG M. L., N GUYEN Q. V.: A space efficient clus-
NOWSKI J., G IEGERICH R.: PathFinder: reconstruction and dy- tered visualization of large graphs. In Proceedings of the 4th
namic visualization of metabolic pathways. Bioinformatics 18, 1 International Conference on Image and Graphics (2007), ICIG
(2002), 124–129. 17 ’07, IEEE, pp. 920–927. 7, 11, 14
[GK07] G ANSNER E. R., KOREN Y.: Improved circular layouts. [Hol06] H OLTEN D.: Hierarchical Edge Bundles: Visualization
In Proceedings of the 15th International Symposium on Graph of adjacency relations in hierarchical data. IEEE Transactions on
Drawing, vol. 4372 of GD ’07. Springer, 2007, pp. 386–398. 13 Visualization and Computer Graphics 12, 5 (2006), 741–748. 5,
[GMT09] G AO J., M ISUE K., TANAKA J.: A multiple-aspects vi- 7, 9, 10, 11
sualization tool for exploring social networks. In Proceedings of [HRD10] H ENRY R ICHE N., DWYER T.: Untangling Euler di-
the Symposium on Human Interface and the Management of In- agrams. IEEE Transactions on Visualization and Computer
formation and Interaction, vol. 5618 of HCI ’09. Springer, 2009, Graphics 16, 6 (2010), 1090–1099. 7, 10, 14, 16
pp. 277–286. 14, 17
[HSS15] H ADLAK S., S CHUMANN H., S CHULZ H.-J.: A sur-
[GSZ∗ 11] G ILBERT F., S IMONETTO P., Z AIDI F., J OURDAN F., vey of multi-faceted graph visualization. In Eurographics Con-
B OURQUI R.: Communities and hierarchical structures in dy- ference on Visualization (EuroVis) - STARs (2015), EuroVis ’15,
namic social networks: analysis and visualization. Social Net- Eurographics Association, pp. 1–20. 4
work Analysis and Mining 1, 2 (2011), 83–95. 17
[IMMS09] I TOH T., M UELDER C. W., M A K.-L., S ESE J.: A
[GZ11] G OU L., Z HANG X.: TreeNetViz: Revealing patterns of hybrid space-filling and force-directed layout method for visu-
networks over tree structures. IEEE Transactions on Visualiza- alizing multiple-category graphs. In Proceedings of the IEEE
tion and Computer Graphics 17, 12 (2011), 2449–2458. 7, 9, 14, Pacific Visualization Symposium (2009), PacificVis ’09, IEEE,
17 pp. 121–128. 6, 7
[HB05] H EER J., B OYD D.: Vizster: visualizing online social [JJ10] JANSSON A., J IRSTRAND M.: Biochemical modeling with
networks. In Proceedings of the 11th IEEE Symposium on Infor- systems biology graphical notation. Drug Discovery Today 15, 9-
mation Visualization (2005), InfoVis ’05, IEEE, pp. 32–39. 17 10 (2010), 365–370. 13, 17
[HBF08] H ENRY N., B EZERIANOS A., F EKETE J.-D.: Improv- [JKS06] J UNKER B., K LUKAS C., S CHREIBER F.: VANTED: A
ing the readability of clustered social networks using node du- system for advanced data analysis and visualization in the context
plication. IEEE Transactions on Visualization and Computer of biological networks. BMC Bioinformatics 7 (2006), 109–121.
Graphics 14, 6 (2008), 1317–1324. 7, 12, 14, 16, 17 17
[HCvW07] H OLTEN D., C ORNELISSEN B., VAN W IJK J. J.:
[JRHT14] J IANU R., RUSU A., H U Y., TAGGART D.: How to
Trace visualization using hierarchical edge bundles and massive
display group information on node-link diagrams: an evaluation.
sequence views. In Proceedings of the 4th IEEE International
IEEE Transactions on Visualization and Computer Graphics 20,
Workshop on Visualizing Software for Understanding and Anal-
11 (2014), 1530–2626. 14, 15, 16, 18
ysis (2007), VISSOFT ’07, IEEE, pp. 47–54. 7, 9
[KFBW14] K URZHALS K., F ISHER B., B URCH M., W EISKOPF
[HD12] H ASCOËT M., D RAGICEVIC P.: Interactive graph match-
D.: Evaluating visual analytics with eye tracking. In Proceed-
ing and visual comparison of graphs and clustered graphs. In Pro-
ings of the Fifth Workshop on Beyond Time and Errors: Novel
ceedings of the International Working Conference on Advanced
Evaluation Methods for Visualization (2014), BELIV ’14, ACM,
Visual Interfaces (2012), AVI ’12, ACM, pp. 522–529. 14, 17
pp. 61–69. 18
[Hea96] H EALEY C. G.: Choosing effective colours for data visu-
[KG06] K UMAR G., G ARLAND M.: Visual exploration of com-
alization. In Proceedings of the 7th Conference on Visualization
plex time-varying graphs. IEEE Transactions on Visualization
(1996), VIS ’96, IEEE Computer Society Press, pp. 263–270. 6,
and Computer Graphics 12, 5 (2006), 805–812. 7, 10, 18
18
[HF06] H ENRY N., F EKETE J.-D.: MatrixExplorer: a dual- [Kit03] K ITANO H.: A graphical notation for biochemical net-
representation system to explore social networks. IEEE Trans- works. BIOSILICO 1, 5 (2003), 169–176. 13, 17
actions on Visualization and Computer Graphics 12, 5 (2006), [KK89] K AMADA T., K AWAI S.: An algorithm for drawing gen-
677–684. 5, 13, 14 eral undirected graphs. Information Processing Letters 31, 1
[HFM07] H ENRY N., F EKETE J.-D., M C G UFFIN M. J.: Node- (1989), 7–15. 5
Trix: a hybrid visualization of social networks. IEEE Trans- [LDB11] L AMBERT A., D UBOIS J., B OURQUI R.: Pathway pre-
actions on Visualization and Computer Graphics 13, 6 (2007), serving representation of metabolic networks. Computer Graph-
1302–1309. 7, 12 ics Forum 30, 3 (2011), 1021–1030. 6

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 23

[Lii10] L IIV I.: Seriation and matrix reordering methods: An projects. In Prooceedings of 6th International Asia-Pacific Sym-
historical overview. Statistical Analysis and Data Mining 3, 2 posium on Visualization (2007), APVIS ’07, IEEE, pp. 25–32.
(2010), 70–91. 5, 12 17
[LPP∗ 06] L EE B., P LAISANT C., PARR C. S., F EKETE J.-D., [PDF14] P ERIN C., D RAGICEVIC P., F EKETE J.-D.: Revisiting
H ENRY N.: Task taxonomy for graph visualization. In Proceed- Bertin matrices: New interactions for crafting tabular visualiza-
ings of 2006 AVI workshop on BEyond time and errors: novel tions. IEEE Transactions on Visualization and Computer Graph-
evaluation methods for information visualization (2006), BELIV ics 20, 12 (2014), 2082–2091. 5
’06, ACM, pp. 1–5. 14, 15 [PF15] PADUANO F., F ORBES A. G.: Extended LineSets: a vi-
[LQB12] L AMBERT A., Q UEYROI F., B OURQUI R.: Visualizing sualization technique for the interactive inspection of biological
patterns in node-link diagrams. In Proceedings of the 16th In- pathways. In Proceedings of the 5th Symposium on Biological
ternational Conference on Information Visualisation (2012), IV Data Visualization (2015), vol. 9 of BioVis ’15, BMC, p. S4. 6,
’12, IEEE, pp. 48–53. 7, 8, 10 7, 10, 14
[LSKS10] L EX A., S TREIT M., K RUIJFF E., S CHMALSTIEG D.: [PGKG08] P INZGER M., G RAFENHAIN K., K NAB P., G ALL
Caleydo: Design and evaluation of a visual analysis framework H. C.: A tool for visual understanding of source code depen-
for gene expression data in its biological context. In Proceedings dencies. In Proceedings of the 16th International Conference on
of the IEEE Pacific Visualization Symposium (2010), PacificVis Program Comprehension (2008), ICPC ’08, IEEE, pp. 254–259.
’10, IEEE, pp. 57–64. 7, 11 17
[PK06] PALEY S. M., K ARP P. D.: The pathway tools cellular
[LWC∗ 14] L IU S., WANG X., C HEN J., Z HU J., G UO B.: Topic-
overview diagram and omics viewer. Nucleic Acids Research 34,
Panorama: a full picture of relevant topics. In Proceedings of the
13 (2006), 3771–3778. 17
Conference on Visual Analytics Science and Technology (2014),
vol. 20 of VAST ’14, IEEE, pp. 183–192. 6, 7, 14 [PLS∗ 13] PARTL C., L EX A., S TREIT M., K ALKOFEN D.,
K ASHOFER K., S CHMALSTIEG D.: enRoute: dynamic path ex-
[LYKB08] L ETUNIC I., YAMADA T., K ANEHISA M., B ORK P.: traction from biological pathway maps for exploring heteroge-
iPath: interactive exploration of biochemical pathways and net- neous experimental datasets. BMC Bioinformatics 14, Suppl 19
works. Trends in Biochemical Sciences 33, 3 (2008), 101–103. (2013), S3. 14, 17
17
[PS06] P ERER A., S HNEIDERMAN B.: Balancing systematic and
[MGK11] M ELVILLE A., G RAHAM M., K ENNEDY J.: Com- flexible exploration of social networks. IEEE Transactions on
bined vs. separate views in matrix-based graph analysis and com- Visualization and Computer Graphics 12, 5 (2006), 693–700. 17
parison. In Proceedings of the 15th International Conference on
Information Visualisation (2011), IV ’11, IEEE, pp. 53–58. 14, [PSK11] P ENG W., S I -K UN L.: Social network analysis layout
16 algorithm under ontology model. Journal of Software 6, 7 (2011),
1321–1328. 14, 17
[MH15] M EIDIANA A., H ONG S.-H.: MultiStory: Visual ana-
[Pur02] P URCHASE H. C.: Metrics for graph drawing aesthetics.
lytics of dynamic multi-relational networks. In Proceedings of
Journal of Visual Languages & Computing 13, 5 (2002), 501–
the IEEE Pacific Visualization Symposium (2015), PacificVis ’15,
516. 2
IEEE, pp. 75–79. 7, 8, 18
[PvW06] P RETORIUS A. J., VAN W IJK J. J.: Visual analysis of
[MML07] M UELLER C., M ARTIN B., L UMSDAINE A.: A com- multivariate state transition graphs. IEEE Transactions on Visu-
parison of vertex ordering algorithms for large graph visualiza- alization and Computer Graphics 12, 5 (2006), 685–692. 7, 9
tion. In Proceedings of the 6th International Asia-Pacific Sympo-
sium on Visualization (2007), APVIS ’07, IEEE, pp. 141–148. 2, [PvW07] P RETORIUS A. J., VAN W IJK J. J.: Bridging the seman-
5, 12 tic gap: Visualizing transition graphs with user-defined diagrams.
IEEE Computer Graphics and Applications 27, 5 (2007), 58–66.
[MZ11] M ISUE K., Z HOU Q.: Drawing semi-bipartite graphs in 17
anchor+matrix style. In Proceedings of the 15th International
Conference on Information Visualisation (2011), IV ’11, IEEE, [PvW08] P RETORIUS A. J., VAN W IJK J. J.: Visual inspection
pp. 26–31. 7, 12 of multivariate graphs. Computer Graphics Forum 27, 3 (2008),
967–974. 9, 13, 14, 17
[NIS15] NAKAZAWA R., I TOH T., S AITO T.: A visualization
[RB10] ROSVALL M., B ERGSTROM C.: Mapping change in large
of research papers based on the topics and citation network. In
networks. PloS one 5, 1 (2010), e8694. 17
Proceedings of the 19th International Conference on Information
Visualisation (2015), IV ’15, IEEE, pp. 283–289. 7, 10 [RFG05] R ATZINGER J., F ISCHER M., G ALL H. C.: EvoLens:
lens-view visualizations of evolution data. In Proceedings of the
[NIST12] NAKAZAWA R., I TOH T., S ESE J., T ERADA A.: In- 8th International Workshop on Principles of Software Evolution
tegrated visualization of gene network and ontology applying a (2005), IWPSE ’05, IEEE, pp. 103–112. 17
hierarchical graph visualization technique. In Proceedings of
the 16th International Conference on Information Visualisation [RHR∗ 10] ROHRSCHNEIDER M., H EINE C., R EICHENBACH
(2012), IV ’12, IEEE, pp. 81–86. 6, 7 A., K ERREN A., S CHEUERMANN G.: A novel grid-based vi-
sualization approach for metabolic networks with advanced fo-
[Noa07] N OACK A.: Energy models for graph clustering. Journal cus&context view. In Proceedings of the 17th International Sym-
of Graph Algorithms and Applications 11, 2 (2007), 453–480. 5 posium on Graph Drawing, vol. 5849 of GD ’09. Springer, 2010,
[NSC05] N EUMANN P., S CHLECHTWEG S., C ARPENDALE S.: pp. 268–279. 7, 11, 12, 17
ArcTrees: Visualizing relations in hierarchical data. In Proceed- [RMF12] RUFIANGE S., M C G UFFIN M. J., F UHRMAN C. P.:
ings of the 7th Joint Eurographics / IEEE VGTC Conference TreeMatrix: A hybrid visualization of compound graphs. Com-
on Visualization (2005), EuroVis ’05, Eurographics Association, puter Graphics Forum 31, 1 (2012), 89–101. 7, 12, 14, 17
pp. 53–60. 7, 9, 11, 14
[RMM15] R ENOUST B., M ELANÇON G., M UNZNER T.: Detan-
[OMB∗ 07] O GAWA M., M A K., B IRD C., D EVANBU P., G OUR - gler: Visual analytics for multiplex networks. Computer Graph-
LEY A.: Visualizing social interaction in open source software ics Forum 34, 3 (2015), 321–330. 13

submitted to COMPUTER GRAPHICS Forum (4/2016).

24 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey

[RPD09] R EITZ F., P OHL M., D IEHL S.: Focused animation of Drawing, vol. 7704 of GD ’12. Springer, 2013, pp. 487–498. 7,
dynamic compound graphs. In Proceedings of the 13th Interna- 8, 18
tional Conference on Information Visualisation (2009), IV ’09,
[SSK14] S AKET B., S IMONETTO P., KOBOUROV S.: Group-
IEEE, pp. 679–684. 7, 11, 18
level graph visualization taxonomy. In EuroVis - Short Papers
[RRAS08] ROYER L., R EIMANN M., A NDREOPOULOS B., (2014), EuroVis ’14, The Eurographics Association. 2, 14, 15,
S CHROEDER M.: Unraveling protein networks with power graph 18
analysis. PLoS Computational Biology 4, 7 (2008), e1000108. 17
[SSKB14] S AKET B., S IMONETTO P., KOBOUROV S. G.,
[RTJ∗ 11] R EDA K., TANTIPATHANANANDH C., J OHNSON A., B ÖRNER K.: Node, node-link, and node-link-group diagrams:
L EIGH J., B ERGER -W OLF T.: Visualizing the evolution of com- An evaluation. IEEE Transactions on Visualization and Com-
munity structures in dynamic social networks. Computer Graph- puter Graphics 20, 12 (2014), 2231–2240. 14, 15
ics Forum 30, 3 (2011), 1061–1070. 17
[SSKB15] S AKET B., S CHEIDEGGER C., KOBOUROV S. G.,
[SA06] S HNEIDERMAN B., A RIS A.: Network visualization by B ÖRNER K.: Map-based visualizations increase recall accuracy
semantic substrates. IEEE Transactions on Visualization and of data. Computer Graphics Forum 34, 3 (2015), 441–450. 14,
Computer Graphics 12, 5 (2006), 733–740. 7, 11 15
[SBG00] S PRENGER T. C., B RUNELLA R., G ROSS M. H.: H- [ST08] S ANTAMARÍA R., T HERÓN R.: Overlapping clustered
BLOB: a hierarchical visual clustering method using implicit sur- graphs: Co-authorship networks visualization. In Proceedings of
faces. In Proceedings of the Conference on Visualization (2000), the 8th International Symposium on Smart Graphics, vol. 5166
VIS ’00, IEEE, pp. 61–68. 7, 10 of SG ’08. Springer, 2008, pp. 190–199. 6, 7, 10, 14
[SCL∗ 09] S HI L., C AO N., L IU S., Q IAN W., TAN L., WANG [SXS∗ 12] S MITH A., X U W., S UN Y., FAEDER J., M ARAI
G., S UN J., L IN C.-Y.: HiMap: Adaptive visualization of large- G. E.: RuleBender: integrated modeling, simulation and visu-
scale online social networks. In Proceedings of the IEEE Pacific alization for rule-based intracellular biochemistry. BMC Bioin-
Visualization Symposium (2009), PacificVis ’09, IEEE, pp. 41– formatics 13, Suppl 8 (2012), S3. 17
48. 17
[SZPM10] S ALLABERRY A., Z AIDI F., P ICH C., M ELANÇON
[SGKS15] S TITZ H., G RATZL S., K RIEGER M., S TREIT M.: G.: Interactive visualization and navigation of web search re-
CloudGazer: A divide-and-conquer approach to monitoring and sults revealing community structures and bridges. In Proceed-
optimizing cloud-based networks. In Proceedings of the IEEE ings of the Graphics Interface Conference (2010), GI ’10, ACM,
Pacific Visualization Symposium (2015), PacificVis ’15, IEEE, pp. 105–112. 7, 11, 12, 14
pp. 175–182. 6, 7, 13
[TE10] T ELEA A., E RSOY O.: Image-based edge bundles: Sim-
[SJSJ05] S ANGAL N., J ORDAN E., S INHA V., JACKSON D.: Us-
plified visualization of large graphs. In Proceedings of the 12th
ing dependency models to manage complex software architec-
Joint Eurographics / IEEE VGTC Conference on Visualization
ture. In Proceedings of the 20th Annual ACM SIGPLAN Con-
(2010), EuroVis ’10, Eurographics Association, pp. 843–852. 13,
ference on Object-Oriented Programming, Systems, Languages,
17
and Applications (2005), OOPSLA ’05, ACM, pp. 167–176. 17
[SJUS08] S CHULZ H.-J., J OHN M., U NGER A., S CHUMANN [TLTC05] T ERMEER M., L ANGE C. F. J., T ELEA A., C HAU -
DRON M. R. V.: Visual exploration of combined architectural
H.: Visual analysis of bipartite biological networks. In Proceed-
ings of the 1st Eurographics Conference on Visual Computing for and metric information. In Proceedings of the 3rd IEEE Interna-
Biomedicine (2008), EG VCBM ’08, Eurographics Association, tional Workshop on Visualizing Software for Understanding and
pp. 135–142. 7, 8, 17 Analysis (2005), VISSOFT ’05, IEEE, pp. 1–6. 6, 7, 17

[SKB∗ 14] S CHREIBER F., K ERREN A., B ÖRNER K., H AGEN [TvDEF09] T HEOCHARIDIS A., VAN D ONGEN S., E NRIGHT
H., Z ECKZER D.: Heterogeneous networks on multiple levels. A. J., F REEMAN T. C.: Network visualization and analysis of
In Multivariate Network Visualization, vol. 8380. Springer, 2014, gene expression data using BioLayout Express(3D). Nature Pro-
pp. 175–206. 7, 11 tocols 4 (2009), 1535–1550. 17

[SKL∗ 14] S TOLPER C. D., K AHNG M., L IN Z., F OERSTER F., [VBAW15] V EHLOW C., B ECK F., AUWÄRTER P., W EISKOPF
G OEL A., S TASKO J., C HAU D. H.: GLO-STIX: Graph-level D.: Visualizing the evolution of communities in dynamic graphs.
operations for specifying techniques and interactive exploration. Computer Graphics Forum 34, 1 (2015), 277–288. 7, 8, 11, 18
IEEE Transactions on Visualization and Computer Graphics 20, [VBSW13] V EHLOW C., B URCH M., S CHMAUDER H.,
12 (2014), 2320–2328. 6, 7 W EISKOPF D.: Radial layered matrix visualization of dynamic
[SLAB15] S ANSEN J., L ALANNE F., AUBER D., B OURQUI R.: graphs. In Proceedings of the 17th International Conference on
Adjasankey: Visualization of huge hierarchical weighted and di- Information Visualisation (2013), IV ’13, IEEE, pp. 51–58. 7, 9
rected graphs. In Proceedings of the 19th International Confer- [VBW15] V EHLOW C., B ECK F., W EISKOPF D.: The state of
ence on Information Visualisation (2015), IV ’15, IEEE, pp. 211– the art in visualizing group structures in graphs. In Eurographics
216. 7, 8, 11 Conference on Visualization - STARs (2015), EuroVis ’15, Euro-
[SLK∗ 09] S TREIT M., L EX A., K ALKUSCH M., Z ATLOUKAL graphics Association, pp. 21–40. 2, 3, 4
K., S CHMALSTIEG D.: Caleydo: Connecting pathways and gene [vdEvW14] VAN DEN E LZEN S., VAN W IJK J. J.: Multivariate
expression. Bioinformatics 25, 20 (2009), 2760–2761. 17 network exploration and presentation: From detail to overview
[SMER06] S HEN Z., M A K.-L., E LIASSI -R AD T.: Visual analy- via selections and aggregations. IEEE Transactions on Visual-
sis of large heterogeneous social networks by semantic and struc- ization and Computer Graphics 20, 12 (2014), 2310–2319. 7,
tural abstraction. IEEE Transactions on Visualization and Com- 8
puter Graphics 12, 6 (2006), 1427–1439. 7, 11 [vH03] VAN H AM F.: Using multilevel call matrices in large soft-
[SMM13] S ALLABERRY A., M UELDER C. W., M A K.-L.: Clus- ware projects. In Proceedings of the 9th IEEE Symposium on In-
tering, visualizing, and navigating for large dynamic graphs. formation Visualization (2003), InfoVis ’03, IEEE, pp. 227–232.
In Proceedings of the 20th International Symposium on Graph 7, 9, 17

submitted to COMPUTER GRAPHICS Forum (4/2016).

 C. Vehlow, F. Beck, & D. Weiskopf / Visualizing Group Structures in Graphs: a Survey 25

[VHK∗ 13] V EHLOW C., H ASENAUER J., K RAMER A., R AUE Lecture Notes in Computer Science. Springer, 2015, pp. 138–147.
A., H UG S., T IMMER J., R ADDE N., T HEIS F. J., W EISKOPF 13
D.: iVUN: interactive visualization of uncertain biochemical re- [ZCCB13] Z HAO J., C OLLINS C., C HEVALIER F., BALAKRISH -
action networks. BMC Bioinformatics 14, S-19 (2013), S2. 17 NAN R.: Interactive exploration of implicit and explicit rela-
[vHSD09] VAN H AM F., S CHULZ H.-J., D IMICCO J. M.: Hon- tions in faceted datasets. IEEE Transactions on Visualization and
eycomb: Visual analysis of large scale social networks. In Computer Graphics 19, 12 (2013), 2080–2089. 7, 11
Proceedings of the 12th International Conference on Human- [Zec10] Z ECKZER D.: Visualizing software entities using a ma-
Computer Interaction, vol. 5727 of INTERACT ’09. Springer, trix layout. In Proceedings of the ACM Symposium on Software
2009, pp. 429–442. 7, 9 Visualization (2010), SoftVis ’10, ACM, pp. 207–208. 13, 16, 17
[VHTW13] V EHLOW C., H ASENAUER J., T HEIS F. J., [ZXQ15] Z HOU H., X U P., Q U H.: Visualization of bipartite
W EISKOPF D.: Visualizing edge-edge relations in graphs. In relations between graphs and sets. Journal of Visualization 18, 2
Proceedings of the IEEE Pacific Visualization Symposium (2013), (2015), 159–172. 7, 8, 14
PacificVis ’09, IEEE, pp. 201–208. 13
[ZXYQ13] Z HOU H., X U P., Y UAN X., Q U H.: Edge bundling
[vHvW04] VAN H AM F., VAN W IJK J. J.: Interactive visualiza- in information visualization. Tsinghua Science and Technology
tion of small world graphs. In Proceedings of the 10th IEEE Sym- 18, 2 (2013), 145–156. 12
posium on Information Visualization (2004), InfoVis ’04, IEEE,
pp. 199–206. 7, 11, 12
[vHW08] VAN H AM F., WATTENBERG M.: Centrality based vi-
sualization of small world graphs. Computer Graphics Forum 27,
3 (2008), 975–982. 6, 7
[VKB∗ 15] V EHLOW C., K AO D. P., B RISTOW M. R., H UNTER
L. E., W EISKOPF D., G ÖRG C.: Visual analysis of biological
data-knowledge networks. BMC Bioinformatics 16, 135 (2015).
14, 17
[vLKS∗ 11] VON L ANDESBERGER T., K UIJPER A., S CHRECK
T., KOHLHAMMER J., VAN W IJK J. J., F EKETE J.-D., F ELL -
NER D. W.: Visual analysis of large graphs: State-of-the-art
and future research challenges. Computer Graphics Forum 30,
6 (2011), 1719–1749. 2
[VPF∗ 14] V IHROVS J., P RUSIS K., F REIVALDS K., RUCEVSKIS
P., K REBS V.: An inverse distance-based potential field function
for overlapping point set visualization. In Proceedings of the
5th International Conference on Information Visualization The-
ory and Applications (2014), IVAPP ’14, SciTePress, pp. 29–38.
7, 10, 14
[VRW13] V EHLOW C., R EINHARDT T., W EISKOPF D.: Visual-
izing fuzzy overlapping communities in networks. IEEE Trans-
actions on Visualization and Computer Graphics 19, 12 (2013),
2486–2495. 6, 7, 11, 14, 17, 18
[War04] WARE C.: Information Visualization: Perception for De-
sign. Morgan Kaufmann Publishers Inc., 2004. 6
[WWY∗ 15] W U Y., W U W., YANG S., YAN Y., Q U H.: Interac-
tive visual summary of major communities in a large network. In
Proceedings of the IEEE Pacific Visualization Symposium (2015),
PacificVis ’15, IEEE, pp. 47–54. 7, 10, 14
[XDC∗ 13] X U P., D U F., C AO N., S HI C., Z HOU H., Q U H.:
Visual analysis of set relations in a graph. Computer Graphics
Forum 32, 3 (2013), 61–70. 6, 7, 9, 14, 17
[XW05] X U R., W UNSCH D.: Survey of clustering algorithms.
IEEE Transactions on Neural Networks 16, 3 (2005), 645–678. 4
[YDG∗ 15] YOGHOURDJIAN V., DWYER T., G ANGE G., K IEF -
FER S., K LEIN K., M ARRIOTT K.: High-quality ultra-compact
grid layout of grouped networks. IEEE Transactions on Visual-
ization and Computer Graphics PP, 99 (2015), 1–1. 7, 10, 13
[YLZ∗ 13] YANG J., L IU Y., Z HANG X., Y UAN X., Z HAO Y.,
BARLOWE S., L IU S.: PIWI: Visually exploring graphs based on
their community structure. IEEE Transactions on Visualization
and Computer Graphics 19, 6 (2013), 1034–1047. 14
[YS15] YAMASHITA T., S AGA R.: Edge bundling in multi-
attributed graphs. In Human Interface and the Management of
Information. Information and Knowledge Design, vol. 9172 of

submitted to COMPUTER GRAPHICS Forum (4/2016).

View publication stats