Content uploaded by Fabian Beck
Author content
All content in this area was uploaded by Fabian Beck on Mar 08, 2019
Content may be subject to copyright.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 1
Word-Sized Graphics for Scientific Texts
Fabian Beck, Member, IEEE Computer Society, and Daniel Weiskopf, Member, IEEE Computer Society
Abstract—Generating visualizations at the size of a word creates dense information representations often called sparklines. The
integration of word-sized graphics into text could avoid additional cognitive load caused by splitting the readers’ attention between
figures and text. In scientific publications, these graphics make statements easier to understand and verify because additional
quantitative information is available where needed. In this work, we perform a literature review to find out how researchers have already
applied such word-sized representations. Illustrating the versatility of the approach, we leverage these representations for reporting
empirical and bibliographic data in three application examples. For interactive Web-based publications, we explore levels of interactivity
and discuss interaction patterns to link visualization and text. We finally call the visualization community to be a pioneer in exploring
new visualization-enriched and interactive publication formats.
Index Terms—Sparklines, word-sized graphics, literature survey, text and visualization, interactive documents, scientific publishing.
F
1 INTRODUCTION
WE, as researchers, publish papers like decades ago—a static
text document enriched with figures and tables—although
Web-based text documents would allow us to embed videos,
3D graphics, and interactive visualizations. Such media could aug-
ment the text with illustrating examples, supportive information,
or quantitative evidence. However, these combinations also require
the readers to jump between the text and the additional artifact.
Psychology researchers investigate this split-attention effect [1] as
part of cognitive load theory [2]: splitting the readers’ attention
could increase their working memory load and thus reduce their
resources to process the actual content [3]. Hence, integrating
the information into a single representation could avoid the split-
attention effect and reduce cognitive load [1], [4].
Tufte [5] suggested sparklines—word-sized graphics—as an
approach that embeds visualizations into text and creates a com-
bined representation. For example, we can easily show a data-
rich stock chart as part of this text ; the reader does
not have to switch to a separate figure to read the information.
The usage of word-sized graphics has become popular within
spreadsheets and tables, significantly boosted by the integration
of sparklines into Microsoft Excel and other software products.
However, their embedding into text documents, especially scien-
tific publications, falls behind and has not yet been sufficiently
explored from a scientific perspective.
This work aims at studying the potential of word-sized graph-
ics to enrich scientific texts. It both explores the state of the art
and demonstrates new solutions. Figure 1 illustrates the envisioned
integration process of text and word-sized graphics and previews
application examples. In particular, our main contributions are:
•We perform a literature review with 140 publications to
evaluate how word-sized visualizations have already been
applied to scientific communication (Section 3).
•We discuss three generalizable application examples
(Figure 1) that leverage the embedding of word-sized
visualizations into the text of a publication (Section 4).
•Fabian Beck is with University of Duisburg-Essen, Germany
E-mail: fabian.beck@wiwinf.uni-due.de.
•Daniel Weiskopf is with VISUS, University of Stuttgart, Germany
E-mail: weiskopf@visus.uni-stuttgart.de.
•We investigate levels of interaction for embedding word-
sized visualizations into interactive text documents and
propose reusable interaction patterns (Section 5).
We conclude our work with calling the visualization community
to action (Section 6): Explore visualization-enriched publication
formats! Establish standards and best practices that support other
communities! We consider our work as a step in this direction.
2 WORD-SIZED GRAPHICS
Tufte [5] defined the term sparklines as “data-intense, design-
simple, word-sized graphics.” Although this definition is quite
broad, the term sparklines is often understood in the narrow sense
of word-sized line charts or word-sized bar charts, the latter
sometimes also referred to as sparkbars. In this work, however,
we consider the broader definition to cover all kinds of word-sized
data visualizations: sparklines could be any data-intense visual
representations at the size of a word. In particular, we use the term
word-sized graphics to even include the coding of information
using icon images. An alternative, very similar term is word-scale
graphics/visualizations, defined by Goffin et al. [6] somewhat
broader, for instance, allowing graphics that span paragraphs.
Their definition also includes emojis [7]—emoticons (e.g., “,”)
or pictures as placeholder for words (e.g., “Know?”)—, which
are wide-spread in contemporary written colloquial language.
But since we focus on formal scientific communication only, we
consider these as out of scope for the current work. Goffin et al. [6]
discern between data-driven and non-data-driven graphics—our
examples are all data-driven, even icon representations encode
defined categorial attributes. Another related term is glyph, in
visualization literature often understood as a “small visual object
that depicts attributes of a data record” [8]. On the one hand, many
word-sized graphics or sparklines can be considered as a special
type of glyph, but the term glyph is broader, not limited to size or
aspect ratio of a word. On the other hand, visualizations like small
bar charts might not be understood as a glyph because they do not
constitute a single visual object, but rather a collection.
Word-sized graphics could be embedded into different me-
dia: text, tables, user interfaces, source code, or even other
visualizations. Since scientific publications primarily use text to
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 2
PDF
WWW
Integration based on word-sized graphics:
Static
Interactive
Publication
Visualization
Text
Examples
(a)
(b)
(c)
Fig. 1. Integration of text and visualization within scientific publications using word-sized graphics. Three generalizable examples demonstrate the
practical applicability and potential of the approach: (a) classifying references with icon graphics, (b) discussing bibliographic data supported by
interactive sparklines, and (c) summarizing eye movement data with word-sized visualizations.
communicate information, our main focus is on the integration of
word-sized graphics into text. We consider them not as part of the
language, but as an augmentation (much like a reference or a foot-
note). Also, their usage in tables and figures is relevant in this work
as a secondary means of communication within publications. In
interactive Web-based documents, borders between tables, figures,
and interactive visualizations blur. Even the graphics integrated
into text could provide interactions, for instance, transforming
them into a larger figure on click. We also discuss these interactive
examples of scientific communication.
3 LITERATURE REVIEW
Despite being described in textbooks and implemented in soft-
ware libraries, word-sized graphics are not necessarily applied
by researchers. Since we want to study the usage of theses
visualizations for scientific communication within texts, our start-
ing point is to evaluate the popularity and usage scenarios in
this kind of application. We aim at answering the following
research question: How are word-sized graphics used in scientific
publications? Hence, we systematically search for examples of
word-sized graphics in the research literature and classify their
form of usage. Relevant instances could be spread across multiple
disciplines—the search should not be limited to visualization or
computer science research.
We are not aware of any other survey on the usage of
word-sized graphics in publications across scientific communities.
Tufte [5] discusses general design considerations for sparklines.
Others explore the interplay of sparklines and Gestalt laws [9],
placement options for word-scale visualizations into text [10],
[11], and the function of word-scale visualizations in docu-
ments [6]. A survey on the state of the art in glyph design is
also related [8].
Please note that we ourselves apply word-sized graphics in the
following to describe and quantify the collected literature. We also
add word-sized graphics to give examples of representations used
in the collected literature; these examples are representatives of a
group of visualizations or replica of the original diagrams slightly
adopted for consistency.
3.1 Methodology and Categorization Scheme
We performed a Google Scholar search with search terms
sparkline and Tufte (retrieved: September 5, 2016). We used
Google Scholar as a search engine because it lists publications
from all disciplines and is one of the most extensive collections
of scientific literature. We used the search term sparkline because
it is—more than word-sized graphics—the accepted and unique
term. We added the term Tufte to filter out noise and reduce the
collection to those publications that pay credit to Tufte.
The search produced a result set of 542 publications. We
excluded 38/542 duplicates, 49/542 non-English publications, and
93/542 non-scientific publications (i.e., publications that were
not published in a scientific context, for instance, user manuals,
patents, talks, or blog posts). For the remaining 362/542 publi-
cations, we tried to retrieve a full document copy (e.g., PDF or
HTML version) and were successful in 316/362 cases. In about half
of these publications (176/316 ), we did not detect any original use of
word-sized graphics, but just a reference to the term sparkline, for
instance, as part of a literature overview; we also did not consider
it as original use if sparkline-like graphics were considerably
larger than the height of a few lines or carried axes and labels.
We document these exclusion decisions in a table that is part of
the supplemental material. The remaining 140/542 publications are
the final set on which the following classification is based. The
literature collection together with the classification is available in
the interactive literature browser SurVis [12]:
http://sparklines-literature.fbeck.com
We visually scanned each document for word-sized graphics.
If the document was too long (e.g., PhD theses or books), we
searched for the term sparkline within in the document. Based
on the detected word-sized graphics, we classified the publication
into the following main usage types (Category type):
•Visualization Technique: The work introduces a visual-
ization technique using word-sized graphics.
•User Study: The work evaluates an approach based on
word-sized graphics or their general usage in a user study.
•Meta: The work broadly discusses considerations for using
word-sized graphics and draws general conclusions.
•Scientific Communication: The work uses word-sized
graphics only to communicate data—the visualization it-
self is no main contribution.
The last type is the one we are most interested in within the
scope of this work. While we briefly review publications of the
other types as well, we focus the following literature analysis
onto the examples assigned to the communication type. Our
paper—both making general observations and using word-size
graphics to communicate results—would be classified as meta and
communication.
We further categorize publications according to the scientific
domain they target (Category domain); this domain does not refer
to the scientific community in which the work was published, but
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 3
Fig. 2. Word cloud summary of keywords assigned to publications; sub-
script numbers and font sizes encode frequencies; usage type keywords
are assigned a unique color: visualization technique ,communica-
tion ,user study , and meta ; word-sized graphics next to every
keyword show the overlap of the respective keyword with the usage
types (e.g., for domain:cs , the yellow bar indicates that about a third
of these are communication publications); created with SurVis [12].
the domain of the data visualized in the word-sized graphics. We
also discern different forms of embedding for word-sized graphics
(Category embedding), for instance, tables, lists, texts, and other
representations. Finally, we classify the visual encoding used in
the word-sized graphics (Category vis), for example, line charts,
bar charts, or spatial encodings.
The classification was the result of an iterative coding process,
refining categories during classifying new publications and reclas-
sifying previous publications when adapting the overall scheme.
We assigned at least one keyword for the categories type,domain,
embedding, and vis to every publication in our collection. Only
in case a publication could not unambiguously be associated with
a single keyword, we assigned multiple keywords per category.
To highlight aspects beyond this classification scheme, some
publications carry additional independent keywords.
3.2 Results
The collected publications per year 21
0(plotted from
2006 to 2016) are a general indicator for the popularity of word-
sized graphics within the scientific community: After Tufte [5]
coined the term in 2006, sparklines had a direct impact as they are
referenced immediately (on average, 9.0 publications per year in
2007 to 2009 ). Over the years, their popularity varied
somewhat but steadily increased (on average, 20.0 publications
per year in 2013 to 2015 ; 2016 is not considered here
because the data might not yet be complete).
Figure 2 quantifies the keywords assigned to the publications
as a word cloud, structured by keyword categories and encoding
keyword frequencies in subscript numbers and font sizes. Type
keywords are highlighted and word-sized graphics encode the
relationship of each keyword to these main discerning types. In
the following, we discuss this classification in detail and use the
keyword categories to structure the analysis. At the end of each
analysis part, we formulate an observation that condenses what we
believe is the central outcome of the respective analysis. Please
note that these statements only reflect our interpretation of the
data, but might not be the only valid conclusion that can be drawn.
3.2.1 Usage Type
Like indicated in Figure 2, a majority of 101/140 publications in
our literature collection uses word-sized graphics as part of a vi-
sualization technique. Some publications only marginally discuss
TABLE 1
Existing examples where word-sized graphics are used in publications
for scientific communication.
Visualization Short Description Domain Embedding
[13] Query event sequence Comp. Sc. Text
[14] Discussion subjects Comp. Sc. List
[15] Participant rating Comp. Sc. Table
[16] Cost across time Defense Text + table
[17] Retention/completion rates Education Table
[18] Levels of suspicious activity Comp. Sc. List
[19] Infection rates Bio-medical Table
[20] Water chemistry species Ecology Table
[21] Transaction costs Ecology Table
[22] Volatilization of hydrocarbon Bio-medical Table
[23] Weights of signaling metrics Bio-medical Table
[24] Phasic pressure traces Bio-medical Table
[25] Criticality distributions Comp. Sc. Table
[26] IO rates Comp. Sc. Table
[27] Score by amount of smoking Bio-medical List
[28] Population statistics by age Sociology Table
[29] Concentration of positives Comp. Sc. Table
[30] Temperature signatures Bio-medical List
[31] Topic frequency in publications History Table
[32] Topic importance trend Comp. Sc. Table
[33] Litter fall Ecology Table
[34] Term likelihood per category Linguistics Text
[35] Web service characteristics Comp. Sc. Table
[36] Answer distributions Bio-medical Table
[37] Attraction schemas Bio-medical List
[38] Participant agreement Comp. Sc. Table
[39] Observed behavior per user Comp. Sc. Text + table
the integration of word-sized graphics. In contrast, such graphics
are in the focus of other publications, for instance, to encode
social networks in a matrix [40], debug electronic circuits with an
interactive tabular representation [41], summarize questionnaire
responses in small multiples [42], show the evolution of keywords
in a tag cloud [43], compare time series in a matrix [44], or
visualize geo-located attributes over time [45]. Further, 15/140 pub-
lications evaluate word-sized graphics in user studies, for instance,
the combination of quantitative and qualitative data in time series
visualizations [46], the effectiveness of a medical data display
based on word-sized graphics [47], the influence of graphics em-
bedding into text onto reading behavior [10], or the performance
of word-size data representations within tables [48]. We classified
5/140 publications as meta, discussing the placement of word-
sized graphics in text [11], user interactions with word-sized
graphics [49], aspect ratios of diagrams and sparklines [50], and
micro visualizations as different means of visualization to augment
texts [51], [52].
In contrast to the rather wide use as part of visualization
approaches, word-sized graphics are less common for communi-
cating scientific data and results (27/140 publications). Table 1 lists
these instances in detail, also providing a graphical classification
of each. While the short description of the visualizations indicates
a great variety of approaches, the graphics have in common that
they only play a subordinate role in the publication. Unlike in the
publications classified as visualization technique, here, word-sized
graphics are just used to communicate some information, but do
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 4
not belong themselves to the core contributions of the work. The
following analysis parts look into more details of these examples.
Observation:While word-sized graphics are regularly applied
within visualization approaches and systems, they are yet rarely
used for communicating data within scientific publications.
3.2.2 Application Domains
The application domains for word-sized graphics are diverse. As
shown in Figure 2, our literature collection covers areas such as
computer science, bio-medical applications, geography, business,
engineering, and more. Among these, computer science and bio-
medical applications are most prominent (38/140 and 23/140 publi-
cations). While the variety of topics is quite high also within these
disciplines, we observed some clusters of publications: for com-
puter science, related to software engineering topics (11/38 publi-
cations), and for bio-medical publications, related to clinical data
dashboards showing patient records (7/23 publications). Also,
notable numbers of publications can be found in geographic ap-
plications (13/140 ), business (12/140 ), and engineering (10/140 ). We
marked 19/140 publications as generic regarding their application
domain because they cover visualization research applicable to
many domains, for instance, user study or meta publications .
For the publications classified as scientific communication,
Table 1 shows a similar distribution of application domains with
many examples in computer science and bio-medical applications.
The short descriptions provide more context and illustrate the
high variance of represented data. We could only detect two
clusters of publications using word-sized graphics for a similar
purpose: one for summarizing the distribution of participants’
answers [15], [36], [38], the other representing topic frequencies
or importance [31], [32]. From an abstracted perspective, the
visualized data in all examples only has in common that it was
measured in some sort of experiment.
Observation:The broad coverage of application domains
shows that word-sized graphics have the potential to be used
within publications in all (data-driven) sciences.
3.2.3 Embedding
Word-sized graphics, due to their small size, can be embedded
into different artifacts. Among the reported use cases in our
collections, their usage as small visualizations in cells of the table
dominates (58/140 publications, cf. Figure 2). Showing them as
small multiples in a list—similar to a table with one column or
row—is a variant (37/140 ). The integration of word-sized graphics
into text, although already proposed by Tufte [5], is less common,
but at least 19/140 publications follow this approach. While
these three basic embeddings work with static and interactive
representations alike, there also exist a number of embeddings
that can be applied specifically within interactive user interfaces:
using word-sized graphics to build a dashboard (17/140 ), augment
a larger visualization (10/140 ), enrich source code in an editor
(4/140 ), or add extra information to a map (3/140 ).
Also among the examples of scientific communication, embed-
dings in tables and lists are prevalent (25/27 examples, cf. Table 1).
As highlighted in bold font in the table, 4/27 publications,
however, use an embedding into text (2/27 in co-occurrence with
table): Adar et al. [13] show temporal event sequences of search
queries integrated into the text. Boehmke [16] visualizes evolving
costs of different categories within text and tables. Potts [34]
shows term likelihood per rating category of movie reviews.
Ying and Robilliard [39] summarize specific behavior of users to
discuss typical user strategies. Comparing the three embeddings
(table , list , and text ), we observe that the ratio of
communication examples (height of bar) is larger among table
representations than for lists and texts; a tabular embedding might
be most straightforward to use in a paper.
Within publications not classified as usage type communica-
tion, further relevant examples of text embedding exist: Brandes
and Nick [40] use word-sized graphics to represent the evolving
relationships of pairs of individuals. For song lyrics, Oh [53]
embeds visualizations above each line that encodes melody and
beat. Tinkelman [54] introduces a word-sized graphic designed
to encode losses and profits of companies and demonstrates
this application within four written statements. From a general
visualization research perspective, Goffin et al. [11] explore the
options for placing word-sized graphics into text and identify
seven cases ranging from in-line placement as practiced in this
work to inter-line placement and overlays. Goffin et al. [10] further
investigate in a user study how these placement options interact
with reading behavior. In most other cases that we classified as
text embedding, word-sized graphics are used in the text in only
few instances or the text embedding just plays a marginal role.
Observation:Word-sized graphics included into tables and
lists are much more common than examples integrated into text,
both in publications describing a visualization technique and those
using them for scientific communication.
3.2.4 Visualization
Finally, we compare the visual encoding used in the word-sized
graphics. Figure 2 clearly shows that most examples relate to sim-
ple line or bar charts (100/140 and 40/140 examples,
with an overlap of 17/140 examples that show both; please note
that we also consider area charts as variants of line charts). Similar
to the original examples provided by Tufte [5], some of them are
enriched with additional markers . Another simple encoding
employs grids to encode, for instance, a sequence of states in
colored cells in one or multiple rows (21/140 ). However,
there also exist approaches demonstrating that word-sized graphics
are not limited to these simple diagram types, for instance, graph-
ics that encode spatial trajectories or densities [55], [56], [57],
stacked quantities that form streams [58], small representations
of boxplots to display statistical distributions [59], [60], [61], or
glyphs that encode multivariate properties [9], [40]. Even parallel
coordinates can be represented [42], and networks in simplified
node-link representations [55], [62] or adjacency matrices [55].
The publication years of the respective publications
show that these alternative representations are rather new, mostly
suggested in the past four years.
In the examples classified as scientific communication, also
line and bar charts dominate (cf. Table 1). In addition to stan-
dard lines , some lines carry markers to indicate specific
points and periods in time or average, minimum, maximum, or
last value ; one example plots multiple lines in a single
chart [28]. Bar charts occur with monochrome bars
and bars colored by category or value ; bars can
be also tailored and carry markers [23]. Two of the
communication examples also use a grid-like visualization
to show temporal sequences of predefined topics [14] or encode
ternary vectors [37]. Abstracting from individual representations,
a commonality among most examples is the encoding of time or
sequence on the horizontal axis and of a measured value on the
vertical axis.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 5
Observation:Line and bar charts are frequently applied in
word-sized graphics, using time or sequence as horizontal and
value as vertical dimension. It is also possible to use other types
of visualizations as word-sized graphics, but this is not yet much
explored for scientific communication.
3.3 Discussion
The literature review provides an impression of the current use of
word-sized graphics in scientific publications. Word-sized graph-
ics are popular in general, but their application for scientific
communication is still quite infrequent. However, we may not
have found all publications, in particular, if the authors were
not aware that they are using a form of visualization referred
to as sparklines or do not acknowledge Edward Tufte. Whereas
our classification scheme is simple and does not require much
interpretation, still in some of the retrieved publications, the classi-
fication was ambiguous. Others might have categorized individual
publications differently or would have used another classification
scheme, potentially resulting in different outcomes of the analysis.
As discussed above, the communication examples found cover
mostly usage in tables and simple line and bar charts. Word-sized
graphics are not much used in practice within the text of scientific
publications to communicate data and results. Reasons for this
gap can be manifold, for instance, (i) word-sized graphics could
be difficult to integrate, (ii) authors were afraid that reviewers or
readers are too conservative regarding their layout preferences of a
scientific publication, or (iii) data is too complex to be represented
as word-sized graphics. In contrast, we believe that the integration
is both feasible and promising in many applications.
4 APPLICATION EXAMPLES
To further explore the usage of word-sized graphics for scientific
communication, we study a number of examples that demonstrate
positive use cases. Since the usage of sparklines in tables and
lists is comparably well established and quite similar to normal
tables and lists, we focus here on examples where visualizations
are directly embedded into text. For text, a good integration is
harder to achieve but, if successful, promises to avoid a split-
attention effect [1].
The three examples we discuss in the following cover a variety
of graphics, from icons for simple classification to visualizations
of rather complex data. Unlike the state of the art applied for scien-
tific communication (cf. Table 1), these examples demonstrate that
word-sized graphics are not limited to line and bar charts. Their
areas of application relate to general topics, like communication of
bibliographic data and experiment results, and could be reused in
a similar way in many domains and publications. They illustrate
one possible solution tested in practical application, but do not
claim to be the best solutions. Whereas we motivate design
considerations and suggestions, they are not backed by empirical
evidence. Extensive user studies would be necessary to come up
with guidelines and more detailed recommendations. Parts of the
discussed examples have already been introduced elsewhere [55],
[63], [64], however, without particularly discussing their use
within texts of scientific publications. For each example, we first
introduce the visualization, then generalize it, and finally discuss
advantages and limitations.
4.1 Classification of References
When discussing related work, authors need to highly condense
information about the referenced publications in one sentence or
a few. Often, it is hard for the reader to follow the discussion
and understand the main differences between the briefly discussed
approaches. This problem is particularly critical for state-of-the-
art reports or other literature surveys where large parts of the
paper consist of content like this. Typically, the authors of such a
publication have already carefully structured and classified the ref-
erenced approaches to give an overview and point out differences.
While the classification of publications might be communicated
through page-filling tables, it requires some searching (and maybe
page flipping) to match the reference in the text with the ones in
the table. Hence, using icons for the classification and embedding
them into the text directly next to the references provides an alter-
native or at least additional encoding. Although debatable whether
these icons should be called word-sized graphics or sparklines, we
consider this example because these icons are a visual encoding
of data—an assignment of publications to categories.
4.1.1 Example
In a survey on visualizing group structures, Vehlow et al. [64]
tested this approach. They first introduced and explained the icons,
here, the ones referring to the type of group structure represented:
“The references are [..] marked with the respective
icons: flat or hierarchical , disjoint or
overlapping .”
Within the discussion of the surveyed publications, these icons
are used to classify the publications, often providing an additional
information that is not described explicitly in the text, for instance:
“Group nodes are connected by visual links if any of
their members are related [CDA*14] , [HN07b,
HN07a] .”
Further icons are applied, such as, for the type of over-
lap , the graph visualization paradigm , the encoding
of time , and the usage of color coding . One of the tables
in the document acts as a legend for the icons in addition to their
introduction in the text. Icons are only added if the classification is
helpful as an additional information in the specific context. Within
table headers and figure captions, icons are used as well to specify
listed approaches.
4.1.2 Generalization
This approach is easy to generalize even if it was designed
for a specific example. In essence, each reference is assigned a
number of categories, each expressed with a different icon. While
it is an important goal to make these icons as self-explaining
as possible, like for acronyms, one should always add a textual
explanation before using the icons for the first time. In context
of visualization literature, finding self-explaining icons might be
easier than in other domains because one can often use stylized
small versions of basic visualization approaches; for instance,
Kucher and Kerren [65] provide another example of such icons
for text visualization. However, other domains have their visual
languages and accepted metaphors that can be used as a basis for
designing the icons. Often, visual identifiers could be borrowed
from icons used in software systems popular in the respective
domain. While it might be possible to use arbitrary icons that are
not obviously related to the category they represent, we do not
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 6
expect that readers would be willing to learn this artificial visual
language—the overhead would more than compensate the gain.
4.1.3 Discussion
Classifying references with icons is a lightweight way of using
word-sized graphics within scientific publications. The encoding
is easy to understand and introduce. The icons do not require much
space or an enlarged version in the paper. The icons only occur
when references, as non-language artifacts, enrich the text with
extra information anyways—the reader is already accustomed to
intermission of reading in this context. As main advantages we see
that the icons (i) show additional illustrations of the referenced
publications, (ii) provide indicators of what are similarities and
differences between referenced publications, and (iii) allow the
reader to quickly search in the document for specific aspects rep-
resented by icons. We recommend focusing on the most important
aspects and leaving out icons if the classification is obvious or
redundant in order to avoid visually overloading the paper.
4.2 Bibliographic Analysis
While the approach discussed above classifies individual refer-
ences qualitatively, we could also consider a more quantitative
approach of discussing and summarizing scientific literature—
counting structured publications in a literature collection and
investigating temporal trends.
4.2.1 Example
You have already seen examples of this use case as part of our
literature search (Section 3). In particular, we use three different
encodings that aggregate publications to quantities:
•Relative Publication Numbers (e.g., 42/140 ): Two over-
lapping bars show the number of publications in relation to
a reference number (i.e., all publications of the collection
or of a previously defined subset). We also provide the
absolute numbers written as a quotient.
•Publication Timelines (e.g., 21
0): Publication
frequencies per year plotted as bar charts give insights
into impact and trends. At first occurrence, we augment
this representation with labels for minimum and maximum
value. Coloring the bars helps us highlight different pe-
riods or subsets of the publications. These timelines are
similar to the line-based visualization of topic frequency
used by Milligan [31].
•Keyword Overlap (e.g., ): Showing the usage types of
a certain publication set identified by a keyword, we illus-
trate the underlying subset structure of these publications.
These graphics were created with the SurVis system [12].
We now show that these can be used also to enrich a textual
description of a literature collection.
4.2.2 Generalization
Since the applied visualizations are simple, they obviously gener-
alize to reflect arbitrary relative quantities, aggregated number of
events or frequencies on a timeline, or overlap of item sets. One
could discuss other document or media collections in a similar
style, for instance, patents, historic texts, works of art, or music.
Another relevant extensions could be to include more metadata
into the discussion, such as information about authors, publishers,
or citations. Whereas some of this information can be represented
with the proposed visualizations already (e.g., citations per year
133
0, 2006–2016 of the book Beautiful Evidence by
Tufte [5]; retrieved from Google Scholar, September 30, 2016),
other data requires new word-sized visualizations (e.g., citation or
co-author networks).
4.2.3 Discussion
In analogy to a reference or a parenthetical explanation, we add
the visualizations usually at the end of the respective sentence or
clause it refers to. However, since relative publication quantities
can be considered as a number, we also embed them within the
sentence like a number. One might argue that our encoding of
publication numbers is superfluous because a single number as
plain text is sufficient. However, we assume that often the absolute
number of publications within a category is important, as well as
the percentage with respect to a total number. If just providing
the absolute number, the reference is not always clear; if just
providing a percentage value, the reader needs to calculate the
absolute number. The graphical representation further provides
the advantage to quickly search for high quantities and compare
subsets by scanning through the quantities.
We pay attention that we do not change the reference of the
graphics within a certain context, that is, the reference number of
the publication quantities, the normalization and time period of
the timelines, and the color coding of the overlap visualizations.
Otherwise, the readers might misinterpret the graphics or each
instance would need to be accompanied with lengthy explanation,
partly destroying the embedding effect.
4.3 Eye Movement Visualizations
Eye tracking is a specialized, but increasingly used empirical
evaluation methodology, for instance, in psychology, human-
computer interaction, and visualization research. Reporting the
results of such studies is difficult because we have to deal
with spatio-temporal patterns of eye movements. While a variety
of visualizations is available for this purpose already [66], an
open research question was if these visualizations could also be
transformed into word-sized representations and stay readable.
Such a representation allows us to also embed eye movement
visualizations into text, which we explore in this example. The
variety and complexity of the visualizations show that the use of
word-sized graphics is not limited to line and bar charts.
4.3.1 Example
In previous work, Beck et al. [55], [63] discuss how to transform
existing eye movement visualizations into word-sized representa-
tions. While demonstrating the usage of the resulting word-sized
visualizations in tabular interfaces, the authors just casually inte-
grated the word-sized visualizations into the text of the publication
for giving an example or performing a case study. In the following,
we discuss this usage scenario in more detail, but first provide a
brief introduction to a selection of the suggested visualizations.
Eye movement data consists of fixations and short transitions
in between them, called saccades. Plotting the spatial position
of these fixations connected with straight lines according to their
sequence of occurrence creates a simple scan path visualization
that can be shown at size of a word (here, a color
map indicates the progress of time). When spatially ag-
gregating the fixations and plotting their density, we obtain an
attention map visualization . Putting time in the focus of
the visualization by using the horizontal axis as a timeline, we
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 7
could plot another metric to the vertical axis, for instance, one
of the spatial dimensions of the fixations . To simplify
the analysis and raise the level of abstraction, analysts identify
areas of interest (AOIs) and transform the fixation sequence to a
sequence of AOIs, which can be visualized on a similar timeline
representation (here, colors and vertical position identify
an AOI). Abstracting further, we transform the sequence of AOIs
into a network of AOI transitions and visualize it as a graph, for
instance, in a node-link representation or an adjacency
matrix .
Each of the presented visualization represents the eye move-
ment for a single participant in a single task. Hence, integrating
them into text is useful for illustrating an example of a typical
fixation sequence or highlighting a specific outlier. In contrast to
traditional eye tracking analysis that focuses on statistically aggre-
gating the data, this integration of word-sized graphics much better
supports a qualitative analysis for studying individual behavior in
detail. Even contrasting a few individuals is possible, for instance,
two participants that show a similar temporal sequence of AOIs
versus another two participants forming a different group [63]:
“Participants in Group (a) found a path from origin
to destination quickly and did not verify it fully
. In Group (b), the participants performed a final
verification more thoroughly .”
If more than two or three instances are required to illustrate a
finding, the word-sized graphics may consume a full line; they
could be treated like a full-line formula and be centered, for
instance, a set of visualizations showing different aspects for the
same participant
.
This representation facilitates enlarging the graphics to some
extent. Of course, the visualizations could also be embedded into
a table. To show details, a few of the examples might be displayed
in regular size enriched with labels as a figure.
4.3.2 Generalization
Although specifically developed for eye movement data, most of
these visualization can be easily applied to other, related data.
First, interaction data is very similar to eye movement data because
interactions, like fixations, can be associated with a timestamp
and screen location [55]. Moreover, eye movements are a special
form of trajectories—hence, the visualizations can also be used
to represent mobility patterns or other spatio-temporal data. Parts
of the proposed visualizations show small networks, and hence,
are applicable to any other kind of network, for instance, software
dependencies or social networks.
4.3.3 Discussion
When representing data at the size of a word, in particular
complex data, visual scalability and clutter quickly becomes an
issue. For the suggested word-sized eye tracking visualizations,
we observe problems like overplotting or coarse data
granularity . The AOI-based visualizations become clut-
tered when adding more AOIs. But a larger visualization does not
always solve the scalability problem: In case of overplotting, the
problem becomes only little better, if at all, when the graphic is
enlarged by a magnitude, for instance, by a factor of nine regarding
the area
.
Instead, we might divide the data into smaller chunks or use vi-
sualizations that abstract the data to a higher level of aggregation.
For the spatial visualizations, another problem is that word-sized
graphics usually have a wide landscape format, not necessarily
similar to the aspect ratio of the stimulus. Also, the visualiza-
tions are too small to have the original stimulus shown in the
background like possible in larger representations. Despite these
issues, however, we were surprised ourselves how many regular
eye tracking visualizations can be transformed to a representation
at the size of a word without losing too much of the perceivable
information and visual scalability. A more detailed discussion of
visual scalability of the word-sized representations is available in
previous work [55], as well an expert review of the different eye
tracking visualizations [63].
5 INTERACTION
Like regular information visualizations, word-sized graphics be-
come more powerful when the users or readers can interact with
them—users might retrieve details on demand, explore relation-
ships through brushing and linking, or adapt the visual encoding to
their needs. There have only been few works yet that discuss inter-
actions with word-sized graphics in detail: In context of debugging
electronic circuits, Frishberg [41] describes the design challenges
and user expectations for interactive sparklines embedded into a
tabular interface. Goffin et al. [49] investigate how word-sized
graphics embedded into text can be interactively transformed from
adocument-centric view (i.e., focusing on the text with word-
sized embeddings) to a visualization-centric view (i.e., focusing on
the visualization, potentially enlarged and interactive). They also
describe intents, techniques, and scope of interactions on word-
sized representations. In other examples, word-sized graphics are
integrated into user interfaces, but interacting with these graphics
is not discussed in detail.
Quite orthogonal to previous work, we discuss and classify
levels of interactivity. We do not focus only on interaction re-
garding the visual representation but also the textual content and
how these representations interact. This is particularly relevant for
scientific communication because of the importance of the text in a
publication. We first describe an interactive example that illustrates
how interaction provides extra value within a text embedding of
word-sized graphics. We later generalize interactions to patterns.
5.1 Example
We extend the timeline representation of our bibliographic analysis
example (see Section 4.2) to an interactive version:
http://biblines-example.fbeck.com
We created a new textual description summarizing literature data
publicly available in context of a survey on dynamic graph visu-
alization [67]. The example combines text, word-sized timelines,
and a regular-size timeline diagram. While Figure 3 shows a static
overview of the example, Figure 4 illustrates the interactions:
•Hovering a word-sized graphic marks the clause it refers
to and also shows the displayed data in the regular-size
timeline as an overlay with blue bars (Figure 4 (a)).
•Clicking on a word-sized graphic makes the overlay
persistent—the previous timeline data is replaced (Fig-
ure 3). The word-sized graphic currently shown enlarged
is marked with a light blue background.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 8
Fig. 3. Interactive Web-based integration of word-sized graphics into text
discussing a bibliographic analysis of a literature collection on dynamic
graph visualization [67].
•Hovering a word in bold font highlights a subset of the
data. If the text refers to a time-independent subset of the
data, this data is plotted as an overlay with blue bars on the
enlarged diagram (Figure 4 (b)). If the text describes a time
period, the respective period is highlighted with a yellow
background in the enlarged diagram and in the word-sized
graphic attached to the respective clause (Figure 4 (c)).
•Hovering a bar in the diagram provides details as a
tooltip dialog and marks the respective year with a yellow
background color, in the enlarged diagram as well as in all
word-sized graphics (Figure 4 (d)).
Hence, after clicking on a word-sized graphic, users may
explore the timelines in detail. When hovering a word-sized
graphic or parts of the text, users can compare the temporal
distributions of two publication sets. Hovering years or textual
descriptions of temporal periods allows users to explore and
compare temporal intervals across different representations. Since
the word-sized graphics are explained by the regular-size diagram,
the text is readable in static form already. However, the interactive
version increases the value of the augmentation because it allows
retrieving details and enables comparison. Hence, readers can use
interactions, but do not have to do it to understand the text.
5.2 Levels of Interactivity
The design space of interactivity for word-sized visualizations
spans, as demonstrated, from static publications with no inter-
actions to highly interactive examples. This span is not only a
difference in quantity of available interactions, but also covers
different qualities and scope of interaction. In particular, we
identified the following three levels:
•No Interaction (Level 0): In their basic usage and as
demonstrated above, word-sized graphics embedded into
text do not need to be interactive at all. If not overly
complex or condensed, the small-scale visualizations could
already provide valuable extra information. However, since
the graphics are usually too small to carry axis labels and
(a)
(b)
(c)
(d)
Fig. 4. Interactive interplay of text, word-sized graphics, and a diagram
when hovering (a) a word-sized graphic, (b) highlighted text representing
a subset of the data, (c) highlighted text representing a time period, or
(d) a bar in the diagram.
captions, authors need to take care to provide sufficient ex-
planations in the text surrounding the word-sized graphics.
•Local Interaction (Level 1): A first step when adding
interaction to word-sized graphics is providing details on
demand, for instance, a tooltip dialog with explanations
or showing an enlarged and labeled version of the graphic.
Also, one could link the text and the graphics interactively;
for example, when hovering a related text fragment or
graphic, the respective other element gets highlighted.
These simple interaction techniques share that their effect
is local, only adding or highlighting information in the
local environment of the graphic.
•Global Interaction (Level 2): Going a step further, global
interactions describe connections beyond the local scope
of a graphic. In the above example, instances of a global
interaction are a consistent linking across multiple word-
sized graphics (cf. hovering a bar in the diagram) or a
connection with an enlarged graphic (cf. hovering a word-
sized graphic). Although arbitrary global connections are
possible, we recommend keeping the scope somewhat
restricted to not overwhelm the user with changes.
Hence, according to this classification, the examples as pro-
vided within the PDF version of the paper are obviously non-
interactive (Level 0). In contrast, the interactive example described
above (cf. Section 5.1) contains local and global interaction—
thus, it is classified as Level 2. Transitions from document to
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 9
Tooltip overlay Text integration Dedicated area
Space-filling
overlay
x
Fig. 5. Design space for showing an enlarged visualization as interactive
details-on-demand of a word-sized graphic.
visualization as described by Goffin et al. [49] applied to a single
graphic are an example of local interactions (Level 1). A Level 2
interaction occurs when more than one graphic is involved in an
interaction—discussed as interaction scope by Goffin et al. [49,
Section 3.4].
5.3 Interaction Patterns
We instantiate the interaction levels as reusable patterns applicable
for scientific communication. To this end, we sketch two patterns,
one for each level. Please note that these patterns are only ex-
amples, other instantiations are possible. Nevertheless, we believe
that these patterns are applicable and useful in many scenarios.
5.3.1 Details-on-Demand Interaction
A straightforward local interaction technique (Level 1) for im-
proving the understandability of word-sized graphics is to provide
details on demands. While some textual extra information showing
labels or numbers would be a basic version of such details, these
details could also contain an enlarged, possibly interactive version
of the word-sized representation. For the placement of the details,
there are the following options, which Figure 5 illustrates:
•Space-Filling Overlay: A screen-filling window or graph-
ical layer containing the details overlays the text.
•Tooltip Overlay: A small tooltip dialog overlays part of
the text next to the referenced element in a way that the
referenced element does not get occluded.
•Text Integration: Details are blended in and displace part
of the text—the text layout changes.
•Dedicated Area: Details are shown in a side bar or any
other screen area that is specifically dedicated to this
purpose and does not overlap the text.
If the details show an enlarged graphic, it should not differ
much from the word-sized representation so that readers immedi-
ate see the congruence. We recommend keeping the aspect ratio
consistent and just adding details such as axes and labels. In case
users want to interact with this enlarged version, showing the
details only when hovering is not sufficient—the users need to
make the details persist on screen (e.g., using click) before they
can interact. Implementing such interactions is possible with all
placement options; only for tooltip overlays, one needs to prevent
that they disappear when the mouse leaves the reference. For touch
interfaces, hovering needs to be replaced by tap or tap-and-hold.
5.3.2 Visualization–Text Interaction
As an example of global interaction (Level 2), we abstract
the interactive example described in Section 5.1 to a generic
approach that we call visualization–text interaction. The focus
of this approach is interactively linking visualizations and text.
We assume—like in the example—that the interactive document
Sparkline Figure
Text
Fig. 6. Possible interactive links between text, sparklines, and figures
implementing an visualization–text interaction approach.
does not only contain text and word-sized graphics (here, called
sparklines for simplification) but also regular-size figures. Hence,
interactive, bidirectional links could be integrated between all
three kinds of representations like illustrated in Figure 6:
•Sparkline–Text Interaction : Linking the word-sized
graphics and the text usually refers to a local interaction
that improves the integration. For instance like in the
example, when hovering on a word-sized representation,
the text fragment it refers to becomes highlighted (⇒). The
other way round (⇐), highlighting could be implemented
in a similar way; in the example, we highlight a period
in the related word-sized graphic when hovering a bold-
font text fragment referring to a time period. Although it
is possible to implement global sparkline–text interactions,
we recommend keeping this link local because otherwise
the close integration of text and word-sized representation
would be partly destroyed.
•Text–Figure Interaction : Text and figure can be
connected in a global interaction: a traditional reference
of a figure in the text (e.g., “Figure 6”) could be made
interactive by highlighting the figure when hovering the
reference or the related text fragment (⇒), and vice versa,
all text discussing the figure when hovering the figure (⇐).
This is particularly useful if the document contains several
figures. Since we use only one figure in our example, we
just implemented a link referencing to different parts of the
figure—when a description of a time period is hovered, the
respective period gets highlighted in the figures.
•Sparkline–Figure Interaction : A set of interactive
representations in word-sized graphics and figures can be
considered as a multi-view visualization. Hence, typical
multi-view interaction approaches like brushing and link-
ing can be applied: Selecting something in a figure could
highlight the respective visual elements in the word-sized
graphics (⇒), in our example, a bar referring to a year.
Regarding the opposite direction (⇐), however, selecting
elements in a word-sized graphic is difficult due to their
small size. Still, select operations might refer to all items
represented in the graphic. In the example, hovering a
word-sized graphic overlays the displayed subset of the
data in the figure.
Through these links, text and visualizations become an in-
tegrated unit, mutually enriching each other. Like in a regular
publication without word-sized graphics or interaction, the domi-
nating usage strategy would still be linearly reading the text and
occasionally using visualizations to better understand. However,
we believe the interactive version has advantages on top of a better
information integration: (i) references between representations
become clearer and easier to trace, (ii) interactive links create
a connected visualization showing different facets of the data,
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 10
and (iii) interaction supports alternative reading strategies, such
as studying figures first and then looking for the respective text.
6 CONCLUSION AND CA LL TO ACTION
This paper discussed the potential of embedding word-sized
graphics within scientific texts. A survey of the state-of-the-art
usage revealed that word-sized graphics are both flexible in what
data they represent and universally applicable across scientific do-
mains. However, authors have just started to use them for scientific
communication. We demonstrated generalizable use cases lever-
aging word-sized graphics in a variety of applications. Interactive
publications could further exploit these kinds of visualizations,
providing details on demand and interactively linking text and
visual content.
Call to Action: We believe that the visualization community
can and should play a leading role in establishing word-sized
graphics and other visualization-based enrichments for scientific
texts. Visualization researchers have the expertise to provide
positive examples and set standards. We call members of the
community to action, in particular, (i) authors to experiment with
better integration of text and visualizations in their papers and
explore the use of interactive publication formats, (ii) reviewers to
be open for new ideas and value experimental publication formats,
and (iii) editors, conference chairs, and publishers to accept
and technically support the submission of publications where
borders between text and graphics blur. Evaluating these practical
experiments, visualization researchers could identify best prac-
tices and establish standards supporting other communities. With
these steps, we hope that the visualization community advances
scientific publications into visualization-enriched texts and other
communities will adapt these practices. If successful, publications
will get more effective as a medium for communicating scientific
results and findings.
ACKNOWLEDGMENTS
We thank Yasett Acurana, Tanja Blascheck, Corinna Vehlow, and
Yuliya Volga for helping design and implementing the word-
sized graphics listed in Section 4. Fabian Beck is indebted to
the Baden-W¨
urttemberg Stiftung for the financial support of this
research project within the Postdoctoral Fellowship for Leading
Early Career Researchers.
REFERENCES
[1] P. Ayres and J. Sweller, “The split-attention principle in multimedia
learning,” in The Cambridge Handbook of Multimedia Learning, R. E.
Mayer, Ed. Cambridge University Press, 2005, pp. 135–146.
[2] J. Sweller, “Implications of cognitive load theory for multimedia learn-
ing,” in The Cambridge Handbook of Multimedia Learning, R. E. Mayer,
Ed. Cambridge University Press, 2005, pp. 19–30.
[3] J. Sweller, J. van Merrienboer, and F. Paas, “Cognitive architecture and
instructional design,” Educational Psychology Review, vol. 10, no. 3, pp.
251–296, 1998.
[4] P. Ginns, “Integrating information: A meta-analysis of the spatial conti-
guity and temporal contiguity effects,” Learning and Instruction, vol. 16,
no. 6, pp. 511–525, 2006.
[5] E. R. Tufte, Beautiful Evidence, 1st ed. Graphics Press, 2006.
[6] P. Goffin, J. Boy, W. Willett, and P. Isenberg, “An exploratory study
of word-scale graphics in data-rich text documents,” IEEE Transactions
on Visualization and Computer Graphics, 2016. [Online]. Available:
https://doi.org/10.1109/TVCG.2016.2618797
[7] L. Lebduska, “Emoji, emoji, what for art thou?” Harlot: A Revealing
Look at the Arts of Persuasion, vol. 1, no. 12, 2014.
[8] R. Borgo, J. Kehrer, D. H. S. Chung, E. Maguire, R. S. Laramee,
H. Hauser, M. Ward, and M. Chen, “Glyph-based visualization: Foun-
dations, design guidelines, techniques and applications,” in Eurographics
State of the Art Reports. Eurographics Association, 2013, pp. 39–63.
[9] U. Brandes, B. Nick, B. Rockstroh, and A. Steffen, “Gestaltlines,”
Computer Graphics Forum, vol. 32, no. 3pt2, pp. 171–180, 2013.
[10] P. Goffin, W. Willett, A. Bezerianos, and P. Isenberg, “Exploring the
effect of word-scale visualizations on reading behavior,” in Proceedings
of the 33rd Annual ACM Conference Extended Abstracts on Human
Factors in Computing Systems. ACM, 2015, pp. 1827–1832.
[11] P. Goffin, W. Willett, J.-D. Fekete, and P. Isenberg, “Exploring the
placement and design of word-scale visualizations,” IEEE Transactions
on Visualization and Computer Graphics, vol. 20, no. 12, pp. 2291–2300,
2014.
[12] F. Beck, S. Koch, and D. Weiskopf, “Visual analysis and dissemination
of scientific literature collections with SurVis,” IEEE Transactions on
Visualization and Computer Graphics, vol. 22, no. 1, pp. 180–189, 2016.
[13] E. Adar, D. S. Weld, B. N. Bershad, and S. S. Gribble, “Why we search:
visualizing and predicting user behavior,” in Proceedings of the 16th
International Conference on World Wide Web. ACM, 2007, pp. 161–
170.
[14] A. T. Baker, “Theoretical and empirical studies of software develop-
ment’s role as a design discipline,” Ph.D. dissertation, California State
University at Long Beach, 2010.
[15] F. Beck, B. Dit, J. Velasco-Madden, D. Weiskopf, and D. Poshyvanyk,
“Rethinking user interfaces for feature location,” in Proceedings of the
2015 IEEE 23rd International Conference on Program Comprehension.
IEEE, 2015, pp. 151–162.
[16] B. C. Boehmke, “Grabbing the air force by the tail: Applying strategic
cost analytics to understand and manage indirect cost behavior,” Ph.D.
dissertation, Air Force Institute of Technology, 2015.
[17] J. Bryer and L. Daniels, “Measuring all students: An alternative method
for retention and completion rates,” in Proceedings of the North East
Association of Institutional Research Annual Conference, 2011.
[18] A. Caglayan, M. Toothaker, D. Drapeau, D. Burke, and G. Eaton, “Be-
havioral analysis of botnets for threat intelligence,” Information Systems
and e-Business Management, vol. 10, no. 4, pp. 491–519, 2012.
[19] K. K. Chui, P. Webb, R. M. Russell, and E. N. Naumova, “Geographic
variations and temporal trends of Salmonella-associated hospitalization
in the US elderly, 1991-2004: A time series analysis of the impact of
HACCP regulation,” BMC Public Health, vol. 9, no. 1, pp. 1–10, 2009.
[20] J. J. Duda, M. M. Beirne, K. Larsen, D. Barry, K. Stenberg, and
M. L. McHenry, “Aquatic ecology of the Elwha River estuary prior to
dam removal,” in Coastal Habitats of the Elwha River Washington—
Biological and Physical Patterns and Processes Prior to Dam Removal.
USGS Scientific Investigations Report, 2011, vol. 5120, pp. 175–223.
[21] D. Garrick, S. M. Whitten, and A. Coggan, “Understanding the evolution
and performance of water markets and allocation policy: A transaction
costs analysis framework,” Ecological Economics, vol. 88, pp. 195–205,
2013.
[22] A. Gillespie, H. Sanei, A. Diochon, B. Ellert, T. Regier, D. Chevrier,
J. Dynes, C. Tarnocai, and E. Gregorich, “Perennially and annually
frozen soil carbon differ in their susceptibility to decomposition: analysis
of subarctic earth hummocks by bioassay, XANES and pyrolysis,” Soil
Biology and Biochemistry, vol. 68, pp. 106–116, 2014.
[23] K. A. Janes, H. C. Reinhardt, and M. B. Yaffe, “Cytokine-induced
signaling networks prioritize dynamic range over signal strength,” Cell,
vol. 135, no. 2, pp. 343–354, 2008.
[24] N. P. Johnson, D. T. Johnson, R. L. Kirkeeide, C. Berry, B. De Bruyne,
W. F. Fearon, K. G. Oldroyd, N. H. Pijls, and K. L. Gould, “Repeatability
of fractional flow reserve despite variations in systemic and coronary
hemodynamics,” JACC: Cardiovascular Interventions, vol. 8, no. 8, pp.
1018–1027, 2015.
[25] A. Jordan, “Evaluating and estimating the WCET criticality metric,”
in Proceedings of the 11th Workshop on Optimizations for DSP and
Embedded Systems. ACM, 2014, pp. 11–18.
[26] S. Kavalanekar, B. Worthington, Q. Zhang, and V. Sharda, “Character-
ization of storage workload traces from production Windows servers,”
in Proceedings of the IEEE International Symposium on Workload
Characterization. IEEE, 2008, pp. 119–128.
[27] I. Lang, E. Gardener, F. A. Huppert, and D. Melzer, “Was John Reid
right? Smoking, class, and pleasure: a population-based cohort study in
England,” Public Health, vol. 121, no. 7, pp. 518–524, 2007.
[28] J. S. Lee and E. N. Waithaka, “The intersections of marginalized
social identities in the transition to adulthood: A demographic
profile,” Emerging Adulthood, 2016. [Online]. Available: https:
//doi.org/10.1177/2167696816659021
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 11
[29] M. Lyra, D. Clarke, H. Morgan, J. Reffin, and D. Weir, “High value
media monitoring with machine learning,” KI – K¨
unstliche Intelligenz,
vol. 27, no. 3, pp. 255–265, 2013.
[30] J. R. Mahan, A. W. Young, and P. Payton, “Continuously monitored
canopy temperature as a proxy for plant water status,” American Journal
of Plant Sciences, vol. 6, no. 14, p. 2287, 2015.
[31] I. Milligan, “Illusionary order: Online databases, optical character recog-
nition, and Canadian history, 1997-2010,” Canadian Historical Review,
vol. 94, no. 4, pp. 540–569, 2013.
[32] S. Neuhaus and T. Zimmermann, “Security trend analysis with CVE
topic models,” in Proceedings of the 21st International Symposium on
Software Reliability Engineering. IEEE, 2010, pp. 111–120.
[33] F. Peterson, J. Sexton, and K. Lajtha, “Scaling litter fall in complex
terrain: A study from the western Cascades Range, Oregon,” Forest
Ecology and Management, vol. 306, pp. 118–127, 2013.
[34] C. Potts, “On the negativity of negation,” in Semantics and Linguistic
Theory, vol. 20, 2010, pp. 636–659.
[35] S. J. Schultheiss, M.-C. M¨
unch, G. D. Andreeva, and G. R¨
atsch, “Per-
sistence and availability of web services in computational biology,” PloS
one, vol. 6, no. 9, p. e24914, 2011.
[36] T. Torsvik, B. Lillebo, and G. Mikkelsen, “Presentation of clinical
laboratory results: an experimental comparison of four visualization
techniques,” Journal of the American Medical Informatics Association,
vol. 20, no. 2, pp. 325–331, 2013.
[37] K. Willadsen and J. Wiles, “Robustness and state-space structure of
boolean gene regulatory models,” Journal of Theoretical Biology, vol.
249, no. 4, pp. 749–765, 2007.
[38] C.-S. Wu, “Designing tangible tabletop interactions to support the fitting
process in modeling biological systems,” Ph.D. dissertation, Georgia
Institute of Technology, 2012.
[39] A. T. Ying and M. P. Robillard, “Selection and presentation practices
for code example summarization,” in Proceedings of the 22nd ACM SIG-
SOFT International Symposium on Foundations of Software Engineering.
ACM, 2014, pp. 460–471.
[40] U. Brandes and B. Nick, “Asymmetric relations in longitudinal social
networks,” IEEE Transactions on Visualization and Computer Graphics,
vol. 17, no. 12, pp. 2283–2290, 2011.
[41] L. D. Frishberg, “Interactive sparklines: A dynamic display of quantita-
tive information,” in CHI’11 Extended Abstracts on Human Factors in
Computing Systems. ACM, 2011, pp. 589–604.
[42] A. Kachkaev, J. Wood, and J. Dykes, “Glyphs for exploring crowd-
sourced subjective survey classification,” Computer Graphics Forum,
vol. 33, no. 3, pp. 311–320, 2014.
[43] B. Lee, N. Henry Riche, A. K. Karlson, and S. Carpendale, “Spark-
Clouds: Visualizing trends in tag clouds,” IEEE Transactions on Visual-
ization and Computer Graphics, vol. 16, no. 6, pp. 1182–1189, 2010.
[44] H. Song, B. Lee, B. Kim, and J. Seo, “DiffMatrix: Matrix-based inter-
active visualization for comparing temporal trends,” in EuroVis - Short
Papers. The Eurographics Association, 2012, pp. 103–107.
[45] C. Turkay, A. Slingsby, H. Hauser, J. Wood, and J. Dykes, “Attribute
signatures: Dynamic visual summaries for analyzing multivariate ge-
ographical data,” IEEE Transactions on Visualization and Computer
Graphics, vol. 20, no. 12, pp. 2033–2042, 2014.
[46] W. Aigner, A. Rind, and S. Hoffmann, “Comparative evaluation of an
interactive time-series visualization that combines quantitative data with
qualitative abstractions,” Computer Graphics Forum, vol. 31, no. 3, pp.
995–1004, 2012.
[47] D. T. Bauer, S. Guerlain, and P. J. Brown, “The design and evaluation of
a graphical display for laboratory data,” Journal of the American Medical
Informatics Association, vol. 17, no. 4, pp. 416–424, 2010.
[48] L. M. Parsons and D. Tinkelman, “Testing the feasibility of small
multiples of sparklines to display semimonthly income statement data,”
International Journal of Accounting Information Systems, vol. 14, no. 1,
pp. 58–76, 2013.
[49] P. Goffin, W. Willett, J.-D. Fekete, and P. Isenberg, “Design consid-
erations for enhancing word-scale visualizations with interaction,” in
Proceedings of IEEE VIS 2015 Posters, 2015.
[50] J. Heer and M. Agrawala, “Multi-scale banking to 45 degrees,” IEEE
Transactions on Visualization and Computer Graphics, vol. 12, no. 5,
pp. 701–708, 2006.
[51] J. Parnow, “Micro Visualizations,” Master’s thesis, Potsdam University
of Applied Sciences, 2015.
[52] J. Parnow and M. D¨
ork, “Micro Visualizations: Data-driven typography
and graphical text enhancement,” in Proceedings of IEEE VIS 2015
Posters, 2015.
[53] J. Oh, “Text visualization of song lyrics,” Course cs448b report at Center
for Computer Research in Music and Acoustics, Stanford University,
2010. [Online]. Available: https://ccrma.stanford.edu/∼jieun5/cs448b/
final/Oh final.pdf
[54] D. Tinkelman, “Increasing the transparency and information content of
financial statements using sparklines,” Available at SSRN 1325998, 2009.
[55] F. Beck, T. Blascheck, T. Ertl, and D. Weiskopf, “Word-sized eye
tracking visualizations,” in Eye Tracking and Visualization. Foundations,
Techniques, and Applications (ETVIS 2015), M. Burch, L. Chuang,
B. Fisher, A. Schmidt, and D. Weiskopf, Eds. Springer, 2016, pp.
113–128.
[56] J. H. Goldberg and J. I. Helfman, “Visual scanpath representation,”
in Proceedings of the 2010 Symposium on Eye-Tracking Research &
Applications. ACM, 2010, pp. 203–210.
[57] C. Perin, R. Vuillemot, and J.-D. Fekete, “SoccerStories: A kick-off
for visual soccer analysis,” IEEE Transactions on Visualization and
Computer Graphics, vol. 19, no. 12, pp. 2506–2515, 2013.
[58] M. Burch, T. Munz, F. Beck, and D. Weiskopf, “Visualizing work pro-
cesses in software engineering with Developer Rivers,” in Proceedings
of the 3rd Working Conference on Software Visualization. IEEE, 2015,
pp. 116–124.
[59] H. Barsnes, M. Vaudel, and L. Martens, “JSparklines: Making tabular
proteomics data come alive,” Proteomics, vol. 15, no. 8, pp. 1428–1431,
2015.
[60] A. Kowarik, B. Meindl, and M. Templ, “sparkTable: Generating graphical
tables for websites and documents with R,” The R Journal, vol. 7, no. 1,
pp. 24–37, 2014.
[61] M. Templ, “Correlation between indicators over time in thematic maps,”
Austrian Journal of Statistics, vol. 41, no. 1, pp. 67–79, 2016.
[62] R. P. Radecki and M. A. Medow, “Cognitive debiasing through sparklines
in clinical data displays,” in Proceedings of the 2007 AMIA Symposium,
vol. 11, 2007, p. 1085.
[63] F. Beck, Y. Acurana, T. Blascheck, R. Netzel, and D. Weiskopf, “An ex-
pert evaluation of word-sized visualizations for analyzing eye movement
data,” in Proceedings of ETVIS 2016. IEEE, 2016.
[64] C. Vehlow, F. Beck, and D. Weiskopf, “Visualizing group structures in
graphs: a survey,” Computer Graphics Forum, 2016. [Online]. Available:
https://doi.org/10.1111/cgf.12872
[65] K. Kucher and A. Kerren, “Text visualization techniques: Taxonomy,
visual survey, and community insights,” in Proceedings of the 2015 IEEE
Pacific Visualization Symposium. IEEE, 2015, pp. 117–121.
[66] T. Blascheck, K. Kurzhals, M. Raschke, M. Burch, D. Weiskopf, and
T. Ertl, “State-of-the-art of visualization for eye tracking data,” in EuroVis
- STARs. Eurographics Association, 2014, pp. 63–82.
[67] F. Beck, M. Burch, S. Diehl, and D. Weiskopf, “A taxonomy and survey
of dynamic graph visualization,” Computer Graphics Forum, 2016.
[Online]. Available: htpps://doi.org/10.1111/cgf.12791
Fabian Beck is assistant professor at the Uni-
versity of Duisburg-Essen, Germany. He re-
ceived the Dr. rer. nat. (PhD) degree in computer
science from the University of Trier, Germany in
2013. His research focuses on methods for visu-
alizing and comparing large and dynamic graphs
and hierarchies, often in the context of software
systems and their evolution. He also investigates
visual analytics systems and the integration of
visualizations into text.
Daniel Weiskopf is a professor at VISUS, Uni-
versity of Stuttgart, Germany. He received his
Dr. rer. nat. (PhD) degree in physics from the
University of T¨
ubingen, Germany (2001), and
the Habilitation degree in computer science at
the University of Stuttgart, Germany (2005). His
research interests include all areas of visualiza-
tion, visual analytics, GPU methods, perception-
oriented computer graphics, and special and
general relativity.