A Review on Visualization Recommendation Strategies

Pawandeep Kaur

and Michael Owonibi

Heinz-Nixdorf Chair for Distributed Information Systems, Friedrich-Schiller-University, Jena, Germany

Keywords: Data Visualization, Visualization Recommendation, Visual Mapping, Review, Survey, Graphic Selection.

Abstract: Choosing the best visualization of a given dataset becomes more and more complex as not only the amount

of data, but also the number of visualization types and the number of potential uses of visualizations grow

tremendously. This challenge has spurred on the research into visualization recommendation systems. The

ultimate aim of such a system is the suggestion of visualizations which provide interesting insights into the

data. It should ideally consider data characteristics, domain knowledge and individual preferences to produce

aesthetically appealing and easy to understand charts. Based on the mentioned factors, we have reviewed in

this paper the state-of-the-art in visualization recommendation systems starting from the earliest attempt made

on this subject. We identify challenges to visualization and visualization recommendation to guide future

research directions.

INTRODUCTION

In this big data era, there has been an increase in the

use

of data visualization tools and techniques as a

means to gain insight in the data. It is a lot easier

understand images than words or numbers because

the ability of human cognition

to detect,

analyze and

interpret patterns, anomalies, texture, distance etc. in

graphics. This makes data visualization an important

tool in exploring, analyzing,

and presenting both the

obvious and less obvious features of data. Visualiza-

tion summarizes

data and presents the most relevant

information in a simple and easy-to-understand

way.

The increasing awareness of the importance of vis-

ualization and the vast diversity in types of data vis-

ualized have led

to the generation of a plethora of

visualization classes. For instance, as of the time

of this

writing, more than 300 different visualiza-

tions are listed on the D3.js site. Given this

plethora

of visualization classes, and the various ways

each

class can be used to show a certain aspect of the

data, and ever increasing visualization (analytics) re-

quirements (e.g. presentation, data quality manage-

ment, trend analysis etc.), individuals are

increas-

ingly faced with the difficulty of deciding which vis-

ualization is most appropriate for their task.

This has

led to the development of visualization recommenda-

tion systems.

According to Vartak et al., (2015), a system

providing visualization recommendation should con

sider factors such as data characteristics, intended

goal of the representation, semantics and domain

knowledge represented in the data, ease of under-

standing and aesthetics, and user preference. In this

paper, we use these factors to review the state-of-the-

art in visualization recommendation. The structure of

this paper is as follows: In Section 2, we introduce

and define some visualization terms and concepts. In

Section 3 we present and categorize visualization rec-

ommendation studies based on the area of their con-

tribution. We identify remaining challenges in Sec-

tion 4 before concluding our paper (Section 5).

2 IMPORTANT CONCEPTS

We would like to introduce some concepts related to

the data visualization creation process, which are used

several times in this paper.

Data attributes are associated with variables and

describe their measurement scales, e.g., quantitative,

categorical, ordinal, nominal etc.

Each data attribute is mapped via a process called

visual mapping to some visual

mark. The visual

marks of scatterplot, e.g., include points, X and

axis etc.

Each visualization consists of different visual

marks with different properties. Points in a scatter-

plot, e.g., have some size, shape or color. Bertin

(1983)

names them visual variables.

266

Kaur P. and Owonibi M.

A Review on Visualization Recommendation Strategies.

DOI: 10.5220/0006175002660273

In Proceedings of the 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2017), pages 266-273

ISBN: 978-989-758-228-8

Visualizations can be classified by their

representa-

tional goals or tasks. Scatterplots, e.g., are relevant

for representing ’correlation’ and ’distribution’.

These goals can be achieved by low-level tasks.

Consider, e.g., a bar chart. To achieve the goal "Com-

parison, one needs to identify the sizes of at least two

bars. Here, “identify” is a low-level task and “size” is

a visual variable.

3 VISUALIZATION

RECOMMENDATION

STRATEGIES

Based on the most distinguishing of the factors identified by

Vartak et al. (2015), we classify approaches to visu-

alization recommendation into four distinct catego-

ries.

These categories are defined according to the

main contribution of their research in providing

techniques,

guidelines or directions that assist in rec-

ommending

visualization.

Data Characteristics Oriented: Studies which fall

in this category recommend visualizations based

on data characteristics.

Task Oriented: Studies that fall under this cate-

gory use the representational goals along with the

data characteristics to recommend visualizations.

Domain Knowledge Oriented: Studies which fall

under this category improve the visualization rec-

ommendation process with domain knowledge.

User Preferences Oriented: Studies which fall un-

der this category gather the information about the

user presentation goals and preferences explicitly

through user interaction with the visualization

system.

3.1 Data Characteristics Oriented

Visualization recommendation research studies in this

category have tried to improve the understanding

the data, of different relationships that exist within the

data and of procedures to represent them. The choice

of variables to represent different

aspects of the same

information can greatly influence

the perception and

understanding of the presented information. There-

fore, the research under this category focuses

on: the

definition of new data dimensions or attributes, the

formalization of the process of visual mapping from

data attributes

to visual marks, and the introduction of

new techniques for visual mapping.

The earliest known study that proposes an automa-

tion of graphical designs was that of Gnanamgari’s

Bharat in

1981. As cited by Bouali et al. ( 2015), Bha-

rat proposed some rules for determining which type

of visualization is appropriate for certain data attrib-

utes. However, their work is based on the limited set

of visualizations available in 1981.

Mackinlay’s APT system (Mackinlay, 1986) pro-

poses to formalize and codify the graphical design

specification to automate the graphics generation

process. This is based on composition algebra, which

consists of basis set and composition operators. Be-

fore applying this algebra, data attributes

need to be

encoded with the respective visual mark

which

should be consistent with the rules

presented in Table

Table 1: Data attributes to visual marks mapping

(Mackinlay, 1986).

Nominal Ordinal Quantitative

Size ─ ● ●

Saturation ─ ● ●

Texture ● ● ●

Color ● ●

Orientation ●

Shape ●

In Composition

Algebra, the basis set encodes data

attributes to visual variables (as in Table 1). Compo-

sitional operators generate different presentations by

composing different basis sets from different data at-

tributes. They compose visualizations by

merging

parts which encode the same information. For exam-

ple, two single axis plots with the dot

visual mark

can be composed to a 2D scatterplot.

Later, the specifications based on Mackinlay’s heu-

ristics were used to develop

a research system called

Polaris (Stolte et al., 2002). These specifications were

then revised into a formal declarative visual language

known as VizQL

(Hanrahan, 2006). The visualiza-

tion software Tableau’s (https://public.tab-

leau.com/s/) “Show Me” module (Mackinlay et al.,

2007) uses

VizQL specifications to automatically

recommend visualizations. When the user selects the

data attributes

of his interest, Show Me uses Tab-

leau’s Visual Mapping

rules (Table 2) to define the

visualization types.

In order to enhance the understandability of the data

and the process of visual encoding, Roth and Mattis

(1990) argued that more structural and semantic infor-

mation about the data which is relevant to the presenta-

tion design should be provided. Therefore, they pro-

posed a richer set of data characterizations, divided into

A Review on Visualization Recommendation Strategies

267

different data domains, to be used by humans or ma-

chines for designing visualizations. It includes original

data measurement scales as by (Mackinlay, 1986),

along with new data descriptors: Spatial (coordinates,

name of the city, etc.), Amount (count and discrete

data), Range (duration). They have identified and

grouped the data domains into coverage, cardinality

and uniqueness. Coverage conveys whether every ele-

ment of a set can be mapped to at least one element of

another set. Cardinality expresses the dependency and

‘within’ relationship between two or more attributes of

the same dataset: one to one, one to many, many to

many. Uniqueness refers to the uniqueness of values

within a set or data column. Their proposed character-

istics are used in SAGE, which is a System for Auto-

matic and Graphical Explanation.

Table 2: Tableau Visual Mapping Rule (Mackinlay et al.

2007).

Pane Type

Field 1

Pane Type

Field 2

Mark Type View Type

C C Text Cross-tab

Qd C Bar Bar view

Qd Cdate Line Line view

Qd Qd Shape Scatterplot

Qi C Gantt Gantt view

Qi Qd Line Line view

Qi Qi Shape Scatter plot

Unlike previous work where researchers seek

knowledge from within the relationship between the

variables of the dataset. Shneiderman’s theory (Shnei-

derman, 1996) has emphasized considering the dataset

as a whole collection and understanding the overall re-

lationship between a single collection (like hierarchical

data) or within different data collections. He has cate-

gorized the data into seven dimensions: 1-dimensional,

2-dimensional, 3-dimensional, multi-dimensional, tem-

poral, tree and network data. This proposal serves as

the basis of the implementation of the TIBCO Spotfire

(Shneiderman, 1999).

In the previously mentioned studies and tools,

visualizations were generated offline by specialists.

‘Many Eyes’ changes this trend and provides a first

known public web site where users may upload data

and

create interactive visualizations collaboratively

(Viegas et al., 2007). In Many Eyes, a visualization

is created by matching a dataset with a visualization

component (or visualization techniques). The list of

visualization components is provided in Figure 1.

Figure 1: Many Eyes Visual Mapping Scheme (Viegas

et al., 2007).

Visualization components are separated by horizontal

lines. Each component consists of some visualiza-

tions which share a

common data schema. When the

user selects some data columns, they are mapped with

the data schema which

is associated to some data vis-

ualization. A data schema is a set of

named, typed

slots. For example: ’T’ in the above table is single col-

umn textual data and ’T+’ means that the dataset has

more than one textual data column. Thus, a treemap

(as in Figure 1) can be

expressed as an ordered set of

textual columns, where

each row in the set describes

the path from the top

of the hierarchy to the leaf

item. The dataset and produced visualization then can

be shared with other users for comments, feedback

and future improvement, thus providing a collabora-

tive workbench for visualization creation.

The popularity of Many Eyes has proved the usa-

bility and ease of access of deploying visualization

software as a web application. Along with that, the

dashboard environment provided by Tableau also be-

came a standard for visualization creation interfaces.

Voyager (Wongsuphasawat et al., 2016) is a recent

visualization recommendation web application based

on the dashboard type environment. Voyager uses

the Compass Recommendation Engine, which sug-

gests visualizations based on the statistical properties

of the

data. The suggestions are produced in the form

of Vega-lite specifications (Satyanarayan et al.,

2017). A Vega-lite specification is a JSON object (see

Figure 2)

that describes a single data source, a mark

type, visual encodings of data variables, key-value,

and data transformations including

filters and aggre-

gate functions. The Compass Recommendation En-

gine first suggests a list of visualizations based on the

univariate summary of each variable in the dataset.

Then the user can exclude or include variables from

the list to focus on a particular variable set of interest.

Similar to the study by (Wongsuphasawat et al.,

2016), recent studies have tried to exploit the statistcal

IVAPP 2017 - International Conference on Information Visualization Theory and Applications

268

Figure 2: Vega-lite JSON Object (Wongsuphasawat et al.,

2016).

characteristics of data as an assistance to visualization

recommendation. VizDeck

(Key et al., 2012) is an-

other such initiative. It automatically recommends

ranked and coordinated visualizations (vizlets) based

on the statistical properties of the data. They adopts

a card game metaphor to organ-ize multiple visuali-

zations into interactive visual dashboard applications.

When user selects the data the system initially pre-

sents the small multiple views of the XY Charts (scat-

terplot or line chart based on the data attributes). Us-

ers interacts with these vizlets while keeping the good

one and discarding the unwanted vizlets. User inter-

action makes a system to learn that which vizlets are

more likely to be effective for a dataset with particular

features. The learned information enhances the sys-

tem’s ability to recommend more suitable visualiza-

tions when provided with similar data in the future.

Vartak et al., (2015) uses statistical methods of

probability distribution, distance matrices and devia-

tions to suggest different views of bar chart and line

chart. Their prototype SEEDB computes a deviation

of the subset of the data in comparison to the whole

dataset. It then recommends those visualizations for

which the underlying data (subset of data) has a high

deviation from the current and normal trends reflected

in the whole dataset. They argue that users find visu-

alizations with high deviations more interesting and

expressive.

As summarized in Table 3, the contributions pro-

vided by the studies in this section can be classified

into four broad areas on the basis of their contribution

towards better visualization recommendation:

 Data properties definition: by providing richer

sets of data dimensions and characterization

 Rule definition: by providing rules, specifications

and schemas to manipulate the data and perform

visual mapping

 Language formalization: by defining the specifi-

cation in system understandable language to auto-

mate the process of visual mapping

 Statistics based: by using statistical and explora-

tory data analytics procedures to recommend vis-

ualization

Table 3: Classification Table.

Categories Studies

Data Properties

SAGE (Roth and Mattis, 1990),

TIBCO Spotfire (Shneiderman,

1996)

Rule Definition

APT (Mackinlay, 1986),

Many Eyes (Viegas et al., 2007)

Language

Formalization

VizQL (Hanrahan, 2006)

Vega-Lite (Satyanarayan et al.,

2017)

Statistics

Voyager (Wongsuphasawat et al.,

2016), VizDeck (Key et al., 2012),

SeeDB (Vartak et al., 2015)

3.2 Task Oriented

Visualization recommendation research studies in this

category have designed different techniques to infer

the representational goal or user’s intentions behind

visualizing the data. Differences in goals can greatly

alter the effectiveness of graphical designs.

Roth and Mattis (1990) were the first to contribute

the idea of instigating the user’s information seek-

ing

goal in the visualization design. In their study

they identified different domain-independent infor-

mation seeking goals, e.g. comparison, distribution,

correlation etc.

Based on some sets of representational goals, a

classification scheme for visualization recommenda-

tion was proposed by Wehrend and Lewis (1990) in

the form of a 2D matrix of “objects” vs “operations”.

In this matrix, “objects” are data attributes, “opera-

tions” are representation goal and cells contain visu-

alization techniques.

According to Kerpedjiev et. al. (1997), visualiza-

tion recommendation can be further enhanced by the

use of domain level tasks. Hence, they proposed a

model (Figure 3) to hierarchically decompose do-

main-specific user’s goals (for the “transportation

scheduling” domain) into common domain independ-

ent goals or representation goals which are further as-

sociated with some graphical actions or operations.

For example, in Figure 3, domain-specific goals like

“know-shortfalls” (which means to know the daily

shortfalls in the goods transported) was decomposed

to tasks which include “know-difference”. In turn,

“know-difference” is associated with “differentiate”

A Review on Visualization Recommendation Strategies

269

which is a high level domain independent task or goal

which acts on data. Actions associated with “differ-

entiate” include “enable-lookup” on value of individ-

ual days and “enable-comparison” on those values.

This approach was applied in the development of Au-

toBrief (Kerpedjiev et. al. 1997) which is a multime-

dia presentation system that assists in data analysis.



Figure 3: Goals and Actions (Kerpedjiev et al., 1997).

In all the previous studies, the user task list was man-

ually created. By introducing advanced linguistic

techniques in the visualization creation process, re-

searchers seek an opportunity to automate the deriva-

tion of the user task from a natural language query.

One such study (Zhou and Feiner, 1998) introduced

visual task taxonomy to automate the process of gain-

ing a high level of presentation intents from the text.

This taxonomy interfaces between high level tasks

(presentation intent) that can be accomplished by low

level visualization techniques (visual action). For ex-

ample, the visual task Focus<?x> implies that visual

techniques such as Enlarge<?x> or Highlight<?x>

could be used to focus attention on ?x. Their taxon-

omy and techniques are implemented in IMPROVISE

(Illustrative Metaphor Production in Reactive Object-

oriented Visual Environments)

3.3 Domain Knowledge Oriented

In the visualization development process, it is im-

portant to first characterize the task and data in the

vocabulary of the problem domain, so that a visual-

ization can fulfill the

requirements of users in any

particular target domain (Munzner, 2009). The ob-

jectives of domain knowledge oriented approaches in-

clude sharing such knowledge among different design-

ers and

end users, and reducing the burden upon us-

ers to acquire knowledge about complex visualiza-

tion techniques. Such approaches are not core tech-

niques

to produce visualization, but they provide as-

sistance

to other techniques for improving the perfor-

mance while

recommending visualizations. The

studies falling into this category deal

with gaining

the domain knowledge from existing

knowledge

sources or creating a new one which further assists

in the visualization recommendation process.

The earliest known domain knowledge oriented

visualization recommendation study is RAVE

(Klumpar et al., 1994). RAVE has been used for the

visualization of in-situ

measurement data captured by

the NASA spacecraft. The user needs to select either

a visualization type or a representational goal from a

provided list. On user selection, RAVE triggers the

visualization technique associated with the entries in

a list and provides the resultant graphics. RAVE’s

knowledge-base contains: (1) a set of

visualization

objects that corresponds the technique that can

create a

specific visualization, (2) a set of rules that corre-

sponds to the selection of one particular visualization

technique, (3) the high level task that visualization

can perform like correlation for scatterplot, (4) the

refinements

that a visualization can accept and (5)

the domain(s) in

which it can be used. For example,

the visualization

object that corresponds to the 2D

scatterplot can satisfy the rule “attribute x is related

attribute y”, can accept zooming and color as

refinements, and can be applied in any domain

where

numeric-valued attributes are compared.

To include semantic abilities in the process of rec-

ommendations, Gilson et al., ( 2008) propose a prag-

matic

approach for automatic generation of visuali-

zations

from domain-specific data available on the

web in the

form of ontologies. They have described

a pipeline

that combines ontology mapping from

three different ontologies. In this approach, a web

page is

first mapped to a “domain ontology”, which

stores the

semantics of the specific subject domain.

The “domain ontology” is then mapped to one

more “ visual representation ontologies”, each of

which captures the semantics of a visualization style

(e.g., treemaps). A “ Semantic bridging ontology”

bridges the information from the two ontologies and

holds

key knowledge about the relationships between

data

entities of the source, the subject domain and

the visual

artifacts of the target visualizations. They

have implemented the visualization pipeline in

a pro-

totype, SemViz which functions end-to-end from

source web page to target visualization.

Building upon somewhat similar grounds, Voigt

et al., (2012) propose a novel approach for

knowledge-assisted, context-aware visualization rec-

ommendation for semantic web data. VISO is a mod-

ular visualization ontology composed of seven differ-

ent modules that provides a vocabulary to annotate

both data sources and visualization components.

GRAPHIC module formalizes knowledge in the do-

main of visualization. DATA module characterizes

the data variables and structure. ACTIVITY module

is concerned with the human aspects of visualization

IVAPP 2017 - International Conference on Information Visualization Theory and Applications

270

i.e. tasks, actions and operations. SYSTEM, USER

and DOMAIN module describes the data and visual-

ization context and the domain information. Based on

the shared knowledge from the different modules, a

recommendation algorithm covers both discovery

and context-aware ranking of suitable graphic repre-

sentations.

3.4 User Preferences Oriented

Here, those visualization recommendation strategies

are grouped which gather users’ intentions explicitly

from their behavior and interactional records while

they communicate with the visualization system.

They are also known as behavior driven studies.

Some studies also use probabilistic and machine

learning techniques to predict the patterns of user

choice from these interactional records.

The first known behavior driven study is from

Gotz and Wen (2009). BDVR (Behaviour Driven

Visualization Recommendation) consists of two dis-

tinct phases: Pattern Detection and Visualization Rec-

ommendation. In the first phase, user behavior while

interacting with the visualization system is analyzed

to find meaningful interaction patterns. These pat-

terns are, e.g., scan, flip, swap and drill-down. In the

second phase, a recommendation engine infers a

user’s intent from these detected patterns. In case of

“scan pattern”, e.g., the user interactively ‘inspects’

values over a series of data. Then he ‘compares’ those

series within themselves or over time. From these in-

tents, visual tasks are inferred which later suggest an

alternative visualization to the user which suits more

accurately than their current visualization selection.

A similar study conducted by Steichen et al.,

(2013), has provided results on accumulating infor-

mation from user eye gaze patterns. They recorded

the interaction of the user with a given visualization

to predict the users' visual tasks, as well as user cog-

nitive abilities, including perceptual speed (a measure

of speed when performing simple tasks), verbal work-

ing memory (a measure of storage and manipulation

capacity of verbal information), and visual working

memory (a measure of storage and manipulation ca-

pacity of visual and spatial information). They have

shown that such characteristics have a significant ef-

fect on task efficiency, user preference and ease of use

with visualization systems. These findings are pre-

sented in view of designing visualization systems that

can adapt to each individual user in real-time.

Growing towards the recommendation of more

user-centric and user adaptive visualization, many

systems have applied machine and probabilistic

learning approaches from the user interactions while

browsing through the recommended visualization as

in the case of VizDeck (Key et al., 2012) as discussed

in section 3.1. A study from Mutlu et al. (2016) used

techniques like collaborative and content based filter-

ing to suggest charts by deriving similarity matrix ac-

cording to the information needs of the user and chart

characteristics. First they have designed a crowd-

source study to obtain personalized scores and tags on

each visualization. Then a multi-dimensional scale is

used to estimate aspects of quality of charts for col-

laborative filtering and a tag vector is used to recom-

mend potentially interesting chart based on content.



4 CHALLENGES AND

RESEARCH DIRECTION

The ultimate aim of all visualization recommendation

systems is the suggestion of visualizations which au-

tomatically provide interesting insights in data. Over

the years, researchers have continually expanded the

set of requirements addressed by their systems to de-

velop more aesthetically appealing and user adaptive

visualizations. One such requirement is to apply an

appropriate technique to score and rank the suggested

visualization according to the data domain and the

user preference. Along with this requirement Vartak

et al. (2015) opined to include factors such as rele-

vance, surprise, non-obviousness, diversity, etc. in the

visualization recommendation process.

The challenges of other visualization domains

such as information visualization, scientific visualiza-

tion etc. also affect the visualization recommendation

process. As investigated by Chen (2005) some of

these include usability of a recommendation, scalabil-

ity, visual thinking and analytics, etc.

Looking at the trends in the visualization studies,

we can see that researchers have acknowledged the

need of more user and domain centric visualization

by providing domain specific knowledge based ap-

proaches. However, at the same time the research in

generic recommender systems (e.g. data character-

stics and statistics oriented) is rapidly developing. As

a result, there is also a challenge of keeping pace with

this in the visualization recommender systems com-

munity. Overall, there is the question of which con-

cepts in these generic recommender systems can be

re-used and how can they be effectively implemented

in visualization recommendation.

Moreover, there is an ongoing investigation into

use of formal languages, standards or ontologies to

describe the structure, and specifications of classes of

visualization, and the different tasks that can be asso-

A Review on Visualization Recommendation Strategies

271

ciated with the classes.

Another challenge is the efficiency of visualiza-

tion recommendation given the growing space of

combinational possibilities of ever increasing data

sizes (rows and column), classes of visualization, and

intended tasks. In addition, there is also the challenge

of effectively incorporating human computer interac-

tions into visualization systems.

Furthermore, some other research studies are in-

vestigating the use of visualization recommendation

in data-driven science, and visual analytics. The list

of research directions/challenges are not exhaustive,

but they are interesting examples of the current and

future research activities.

5 CONCLUSIONS

Visualization is becoming an increasingly more im-

portant tool for getting insights into the ever bigger

and more complex data being generated in this era.

As a result, different kinds of visualizations with dif-

ferent characteristics are constantly being developed.

Consequently, deciding which visualization best suits

a user’s data and intention becomes more and more

complex. Visualization recommendation systems at-

tempt to support the user in the decision making. In

this paper, we have discussed research on this topic

has gone through several phases beginning from only

considering the data and chart characteristics to now

where several other factors such as individual prefer-

ences, insight tasks, and domain knowledge are con-

sidered in varying degrees. Still, there is strong need

for additional research in particular to keep the visu-

alization, visualization recommendation and recom-

mender system communities synchronized.

ACKNOWLEDGEMENTS

The work has been funded by the DFG Priority Pro-

gram 1374 "Infrastructure-Biodiversity-Explorato-

ries" (KO 2209 / 12-2).

REFERENCES

Bertin, J., 1983. Semiology of graphics: diagrams, net-

works, maps. University of

Wisconsin Press.

Bouali, F., Guettala, A. and Venturini, G., 2015. VizAssist:

an interactive user assistant for visual data mining. The

Visual Computer.(pp.1-17).

Chen, C., 2005. Top 10 unsolved information visualization

problems. IEEE computer graphics and applications,

25(4).(pp.12-16).

Gilson, O., Silva, N., Grant, P.W. and Chen, M., 2008, May.

From web data to visualization via ontology mapping.

In Computer Graphics Forum, 27(3).(pp. 959-966).

Blackwell Publishing Ltd.

Gotz, D. and Wen, Z., 2009, February. Behavior-driven vis-

ualization recommendation. In Proceedings of the 14th

international conference on Intelligent user interfaces

(pp. 315-324). ACM.

Hanrahan, P., 2006, June. Vizql: a language for query, anal-

ysis and visualization. In Proceedings of the 2006 ACM

SIGMOD international conference on Management of

data (pp. 721-721). ACM.

Kerpedjiev, S., Carenini, G., Roth, S.F. and Moore, J.D.,

1997. AutoBrief: a multimedia presentation system for

assisting data analysis. Computer Standards & Inter-

faces, 18(6).(pp.583-593).

Key, A., Howe, B., Perry, D. and Aragon, C., 2012, May.

VizDeck: self-organizing dashboards for visual analytics. In

Proceedings of the 2012 ACM SIGMOD International Con-

ference on Management of Data.(pp. 681-684). ACM.

Klumpar, D.M, Anderson, K., and Simoudis, A. (1994).

Rave:

Rapid visualization environment. The 1994 God-

dard Conference on Space Applications of Artificial In-

telligence.(pp 29-38).

Mackinlay, J., 1986. Automating the design of graphical

presentations of relational information. ACM Transac-

tions On Graphics (Tog), 5(2).(pp.110-141).

Mackinlay, J., Hanrahan, P. and Stolte, C., 2007. Show me:

Automatic presentation for visual analysis. IEEE Trans-

actions on Visualization and Computer Graphics,

13(6).(pp.1137-1144).

Munzner, T., 2009. A nested model for visualization design

and validation. IEEE transactions on visualization and

computer graphics, 15(6).(pp.921-928).

Mutlu, B., Veas, E. and Trattner, C., 2016. VizRec: Recom-

mending Personalized Visualizations. ACM Transac-

tions on Interactive Intelligent Systems (TiiS),

6(4).(pp.31).

Roth, S.F. and Mattis, J., 1990. Data characterization for

intelligent graphics presentation. In Proceedings of the

SIGCHI Conference on Human Factors in Computing

Systems. (pp. 193-200). ACM.

Satyanarayan, A., Moritz, D., Wongsuphasawat, K. and

Heer, J., 2017. Vega-lite: A grammar of interactive

graphics. IEEE Transactions on Visualization & Com-

puter Graphics, (1).(pp.341-350).

Steichen, B., Carenini, G. and Conati, C., 2013, User-adap-

tive information visualization: using eye gaze data to

infer visualization tasks and user cognitive abilities. In

Proceedings of the 2013 international conference on

Intelligent user interfaces (pp. 317-328). ACM.

Shneiderman, B., 1996, September. The eyes have it: A task

by data type taxonomy for information visualizations.

In Visual Languages, 1996. Proceedings., IEEE Sym-

posium (pp. 336-343). IEEE.

Shneiderman, B., 1999. Dynamic queries, starfield dis-

plays, and the path to Spotfire.

Stolte, C., Tang, D. and Hanrahan, P., 2002. Polaris: A sys-

IVAPP 2017 - International Conference on Information Visualization Theory and Applications

272

tem for query, analysis, and visualization of multidi-

mensional relational databases. IEEE Transactions on

Visualization and Computer Graphics, 8(1).(pp.52-65).

Vartak, M., Huang, S., Siddiqui, T., Madden, S. and

Parameswaran, A., 2015 Towards Visualization Recom-

mendation Systems.Workshop on Data Systems.

for Interactive Analytics (DSIA).

Viegas, F.B., Wattenberg, M., Van Ham, F., Kriss, J. and

McKeon, M., 2007. Manyeyes: a site for visualization at

internet scale. IEEE transactions on visualization and

computer graphics, 13(6),(pp.1121-1128).

Voigt, M., Pietschmann, S. and Meißner, K., 2012,

February. Towards a semantics-based, end-user-cen-

tered information visualization process. In Proc. of the

3rd international workshop on semantic models for

adaptive interactive systems (SEMAIS 2012).

Wehrend, S. and Lewis, C., 1990,. A problem-oriented clas-

sification of visualization techniques. In Proceedings of

the 1st Conference on Visualization'90. (pp. 139-143).

IEEE Computer Society Press.

Wongsuphasawat, K., Moritz, D., Anand, A., Mackinlay, J.,

Howe, B. and Heer, J., 2016. Voyager: Exploratory

analysis via faceted browsing of visualization recom-

mendations. IEEE transactions on visualization and

computer graphics, 22(1), (pp.649-658).

Zhou, M.X. and Feiner, S.K., 1998, January. Visual task

characterization for automated visual discourse syn-

thesis. In Proceedings of the SIGCHI conference on

Human factors in computing systems. (pp. 392-399).

ACM Press/Addison-Wesley Publishing Co.

A Review on Visualization Recommendation Strategies

273