1 | MOTIVATION AND RESEARCH OBJECTIVES
Several ground‐breaking advances in geospatial technologies (e.g., small GPS‐enabled devices and high‐resolution re-mote sensors) have led to an unprecedented abundance of geodata (Sui, Goodchild, & Elwood, 2013). While this pres-ents an opportunity to increase our knowledge and understanding of natural and man‐made processes, it also prespres-ents a challenge for analysts who need to make sense of increasingly large, heterogeneous, and multivariate geodata sets. R E V I E W A R T I C L E
Enabling collaborative GeoVisual analytics:
Systems, techniques, and research challenges
Gustavo Adolfo García‐Chapeton | Frank Olaf Ostermann | Rolf
A. de By | Menno‐Jan Kraak
This is an open access article under the terms of the Creative Commons Attribution‐Non Commercial‐NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non‐commercial and no modifications or adaptations are made.
© 2018 The Authors. Transactions in GIS published by John Wiley & Sons Ltd. Faculty of Geo‐Information Science and Earth
Observation (ITC), University of Twente, Enschede, The Netherlands
Correspondence
Gustavo Adolfo García‐Chapeton, Faculty of Geo‐Information Science and Earth Observation (ITC), University of Twente, 7514AE Enschede, The Netherlands. Email: g.a.garciachapeton@utwente.nl
Abstract
Collaboration across disciplines is recognized as one of the great challenges for research in visual analysis of geographic informa-tion (GeoVisual Analytics, GVA). Considering the increasing availability of geodata and the complexity of analytical prob-lems, the need to advance the support for collaborative work is becoming more pressing and prominent. This article contrib-utes to this objective by reviewing the state‐of‐the‐art of the support for collaborative work in GVA systems and by identify-ing research challenges and proposidentify-ing strategies to address them. We conducted a systematic review, resulting in the identi-fication of 13 collaborative systems, 6 distinct collaborative techniques, and 3 research challenges. We conclude that GVA is moving toward more effective support of multidisciplinary and cross‐domain collaborative analysis. However, to materialize this potential, research is needed to improve the support for hy-brid collaborative scenarios, cross‐device collaboration, and support for time‐critical and long‐term analysis.
The analysis of such datasets is a complex task that benefits from the combination of human and computer analysis capabilities. Computers can store and process large datasets to identify patterns, trends, and outliers. However, unless guided by theory and domain knowledge, such “blind” data mining is likely to produce spurious correlations and meaning-less results (Keim, Kohlhammer, Ellis, & Mansmann, 2010; Thomas & Cook, 2005). Therefore, human skills are needed to formulate hypotheses, parameterize algorithms, select evidence, validate and synthesize results, draw conclusions, and ultimately make decisions (Scheider, Ostermann, & Adams, 2017). GeoVisual Analytics (GVA) enables this combined in-telligence by producing synergy between human analytical skills and computer storage and processing power, to support effective and multidisciplinary understanding, reasoning, and decision‐making on the basis of large and complex geodata sets (Andrienko et al., 2007; Roth & MacEachren, 2016). Recent examples of GVA applications include analysis of crimi-nal activity (Roth, Ross, & MacEachren, 2015), Twitter data (Nelson, Quinn, Swedberg, Chu, & MacEachren, 2015), road accident accumulation zones (Ramos, Silva, Santos, & Pires, 2015), and movement data (Andrienko & Andrienko, 2013).
Many analytical problems are complex, ill‐defined, and broad in scope (Andrienko et al., 2007; Isenberg et al., 2011; Thomas & Cook, 2005). In this regard, researchers from different domains agree that such analytical prob-lems will benefit from approaches and tools that support reproducible, multidisciplinary collaborative work (Haklay, 2013; Heer, Viégas, & Wattenberg, 2007; Hey, Tansley, & Tolle, 2009; Isenberg et al., 2011; Ostermann & Granell, 2017; Yovcheva, van Elzakker, & Köbben, 2013). However, many GVA systems are single‐user environments or offer limited support for collaborative work, and in consequence collaboration remains a challenge for GVA research (Andrienko et al., 2007; Çöltekin, Bleisch, Andrienko, & Dykes, 2017; Elmqvist, 2014; Heer & Shneiderman, 2012; Isenberg et al., 2011; Keim et al., 2010; Thomas & Cook, 2005).
To the best of our knowledge, no recent study has investigated the state‐of‐the‐art of the support for collabo-rative work in GVA systems. We address this gap by conducting a systematic review with the following objectives:
1. To identify GVA systems that support collaborative work, and to describe their characteristics regarding
collaborative scenarios and technological platform.
2. To identify and describe collaborative techniques implemented in GVA systems.
3. To identify research challenges to effectively support collaborative work in GVA and propose strategies to
ad-dress these.
This review follows the guidelines for systematic reviews proposed by Kitchenham and Charters (2007). These guide-lines were originally designed for the field of software engineering, but have been adopted successfully in other domains such as information visualization (Yusoff & Salim, 2015), spatio‐temporal analysis (Steiger, de Albuquerque, & Zipf, 2015), and educational resources (Arimoto & Barbosa, 2012). A systematic review has three phases: planning, conducting, and reporting. The planning phase (Sections 1 and 2 of this article) defines the objective of the review, the process to identify the information sources, and the information that will be obtained from them. The conducting phase (Section 2) includes the acquisition of information sources, and the extraction, organization, and synthesis of the information. Finally, the reporting phase (Section 3) prepares a comprehensive document with the review results. Based on these results, we identify research challenges and strategies to address them (Section 4), and draw conclusions about the status of the support for collaborative work in GVA (Section 5).
2 | SYSTEMATIC REVIEW OF SYSTEMS AND COLL ABOR ATIVE
TECHNIQUES
2.1 | Information sources
To identify relevant literature, we used popular electronic databases in the domain of geoscience. The results are
documented in Table 1. The common search and selection criteria (modified to fit each database's query format1)
• Search keywords: (collaborative OR cooperative) AND (“geovisual analytics” OR “geospatial visual analytics” OR geoanalytics)
• Inclusion criteria
• Publication date: between January 2004 and June 2017 inclusive • Publication type: journals, proceedings, transactions, and book chapters • Article type: full text and reviews
• Language: English • Exclusion criteria
• Duplicated papers (identified using EndNote X8)
• Non‐relevant papers (determined by manual paper screening)
To define the search keywords, we investigated the terms used by authors when referring to GVA systems, and con-cluded that commonly used terms are: “GeoVisual Analytics” (e.g., Andrienko et al., 2007), “GeoSpatial Visual Analytics” (e.g., De Amicis, Conti, Piffer, & Simöes, 2009), and “GeoAnalytics” (e.g., Jern, 2009). Given our interest in systems sup-porting analysis performed by teams, we added “Collaborative.” Additionally, given that electronic collaborative systems are based on the principles of computer‐supported cooperative work, we included the keyword “Cooperative.” The cho-sen period marks the introduction of visual analytics (VA) as a research field in 2004 (Wong & Thomas, 2004), and the start of this systematic review.
We defined exclusion criteria to retain only the papers that relate to our research objectives, which are those describing collaborative GVA (CGVA) systems, collaborative techniques, and research challenges for CGVA. For ex-ample, we excluded papers that use the term “collaborative” not in relation to a system, but as general collaborative effort or work.
To include relevant CGVA systems not featured in the academic literature, we extended our search to the web using the Google search engine (www.google.com). After some iterative refinement, the search query was: “collab-oration AND ‘visual analytics software’ ‐paper ‐book ‐conference.” To target active projects only, the search was limited to the last two years (from July 1, 2015 to June 30, 2017). We did not include the term “Geo” in the query be-cause, despite being used with geodata, some systems might not mention it explicitly enough to appear in the search
results. The search produced 56 results,2 of which three systems (SAP BusinessObjects, Oracle BI Visual Analytics,
and SAS Visual Analytics) were not mentioned in the results from the previous search phase.
TA B L E 1 Electronic databases used to identify papers for the systematic review
Source URL Papers Unique results
After paper screening ACM digital library dl.acm.org 19 11 4 GeoBase www.engineeringvillage.com 3 3 2 IEEEXplore ieeexplore.ieee.org 10 10 2
Science direct www.sciencedirect.com 12 12 3
Scopus www.scopus.com 32 22 7
Springer Link link.springer.com 39 34 6
Web of Science apps.webofknowledge.com 9 6 4
Total — 124 99 28
The column “Papers” accounts for the total of papers identified in a database, the column “Unique results” accounts for non‐duplicated papers, and the column “After paper screening” accounts for papers that contribute to address the review objectives.
The multidisciplinary nature of GVA research complicates a fully comprehensive literature review. However, we are confident that our search approach is sufficiently exhaustive to include the most relevant information. Our focus is on the collaborative capacity of the systems based on the supported collaborative scenarios, technological platform, and collaborative techniques. We acknowledge that other perspectives might be adopted, and we encour-age considering them for future research. Given that GVA is a fast‐evolving field, a similar study will yield different search results in the future. However, we argue that extracted major research challenges will persist, and thus en-sure that our findings remain relevant for a longer period. To support comparison with future studies, we adopted a theoretical framework on information extraction and organization, as described in the following subsection.
2.2 | Information extraction and organization
To structure and organize our investigation, we used the knowledge generation model for VA (KGM‐VA) proposed by Sacha et al. (2014) as theoretical framework. This model explicitly separates the human and computer compo-nents to highlight their role in VA, and incorporates the notion of analytical process, as shown in Figure 1. The model provides a clear theoretical separation and order for the stages in the analysis process, although in actual analysis processes the stages may overlap and analysts move back and forth in a dynamic knowledge generation process (Sacha et al., 2014).
The KGM‐VA models the analysis process with three stages termed loops: exploration, verification, and knowl-edge generation. These loops occur in the system's human component, while the computer component provides storage and processing power to support them. The exploration loop represents the interactions of the analysts with the system, which produce findings (i.e., relevant observations about the phenomenon under study). The analysts’ actions are guided by an analytical goal, or in its absence with the aim of defining one. The verification loop guides the exploration loop to confirm hypotheses or to generate new ones. In this loop, the analysts gain insights as the findings are interpreted in the context of the analysis domain, and may contribute to verify or falsify a hypothesis. Finally, in the knowledge generation loop, the analysts combine their expertise with the identified evidence to accept or reject a hypothesis, and generate new knowledge (or suggest further analysis if evidence is not conclusive).
F I G U R E 1 The knowledge generation model for visual analytics explicitly separates systems into human and
computer components, and conceptualizes the analysis process with three loops: exploration, verification, and knowledge generation. Illustration based on Sacha et al. (2014)
Findings Ac on Insight Hypothesis Knowledge Storage and processing
Human component
Device s (visual in te rfaces )Computer component
Explora onLoop Verifica onLoop
Knowledge Genera on
Loop
For our research, the KGM‐VA provides an effective framework to analyze and describe the human and com-puter components and their interaction (system level), and the role of a technique in the analysis process to enable collaboration among participants (technique level).
At system level, our interest was to describe the organization of the human component (i.e., the analysts) to ad-dress the collaborative effort, the provision of computer storage and processing power to support the collaborative effort, and how they are both linked.
A commonly used approach to characterize the organization of participants in a collaborative effort is to consider the time and space in which participation takes place, usually distinguishing four collaborative scenarios: synchro-nous co‐located (same time and space), synchrosynchro-nous distributed (same time and different space), asynchrosynchro-nous co‐ located (different time and same space), and asynchronous distributed (different time and space) (Johansen, 1988). These scenarios are not mutually exclusive and the support for multiple scenarios (or hybrid scenarios) is a desirable characteristic (Isenberg et al., 2011), because it allows a more flexible collaborative workflow. Additionally, a collab-orative effort can also be characterized based on its duration. In this case, two scenarios can be defined: time‐critical (or short‐term) and long‐term. Having a well‐defined set of scenarios allowed us to identify the existence of patterns regarding the organization of the participants.
The collaborative effort is enabled by information and communication technologies. Here, we focus on two de-fining characteristics: first, the provision of storage and processing power by reviewing the deployment options for the system; and second, the supported devices because they link the system's human and machine components.
At technique level, our interest was to describe the defining characteristics of the techniques and describe simi-larity and co‐occurrence of techniques. This allowed us to understand why some techniques are more popular than others, and to identify patterns regarding their roles in the analysis process and combination with other techniques.
The collaborative techniques constitute the mechanism for the analyst to externalize and communicate find-ings and insights with other analysts, which occurs within and across the loops of the analysis process, and enables knowledge generation. This process of collaborative knowledge generation is grounded in the theory of distributed cognition (DC), which considers cognition as a social process, involving human actors as thinking entities and arte-facts as means for knowledge exchange and shared memory (Hollan, Hutchins, & Kirsh, 2000).
3 | RESULTS AND DISCUSSION
From the literature, we identified 10 CGVA systems, complemented by 3 systems from the Google search engine. Six distinct collaborative techniques were extracted from the literature and specific information about the identified systems. In this section, we describe the CGVA systems and collaborative techniques based on the criteria outlined in Section 2.2.
3.1 | Systems
The systematic review identified 13 GVA systems with collaborative features; a summary is presented in Table 2. In this section, we analyze the collaborative scenarios supported by the systems, looking at the space and time charac-teristics of collaboration, as well as the duration of the analysis. Later, we describe the provision of computer storage and processing power to support the collaborative effort. Lastly, we analyze the supported devices as they are vital to link the human and computer components of the system, and to enable collaboration.
In a collaborative system, participants may interact either at the same (synchronous) or a different (asynchro-nous) moment in time, and at the same (co‐located) or a different (distributed) location, which results in four dif-ferent collaborative scenarios (Johansen, 1988). Figure 2 shows those scenarios and the identified systems that support them. The most commonly supported scenario is asynchronous distributed (85% of the systems). This sce-nario is popular because it promotes participation by eliminating the constraints for analysts to synchronize in time
T A B LE 2 Su m m ar y o f c ol lab or at iv e G eo V is ua l an al yt ic s s ys te m s Sy st em na m e Yea r D es cr ip tio n C ol la bor at iv e sc ena rio A pp lic at io ns Re fer en ce s O R A C LE B I V is ua l A na ly tic s U nk no wn Co m mer ci al bu si ne ss in te lli genc e VA s ys te m , a va ila bl e a s c lo ud ‐ ba se d s er vi ce ; s up po rt s m ul tip le de vi ce s A sy nch ro no us di st rib ut ed B us in ess in te lli ge nc e O ra cl e (2 017 ) SA P B us in ess O bj ec ts U nk no wn Co m mer ci al bu si ne ss in te lli genc e VA s ys te m ; s up po rt s m ul tip le de vi ce s, i nc lu di ng a s pe ci al s et up fo r l ar ge s cr ee ns Sy nc hr on ou s c o‐ lo ca te d a nd as yn ch ro no us di st rib ut ed B us in ess in te lli ge nc e SA P (2 01 7) ReV is e 20 06 D es kto p‐ ba se d p ro to ty pe ba se d o n th e I m pr ov is e T oo lk it; im pl emen ts th e m eth od “R e‐ vi su ali za tio n, ” w hi ch a llo w s u se rs t o g en er at e a nd re vi ew a na ly si s s es si on lo gs A sy nc hr on ou s co ‐l oc ate d A na ly si s o f U .S . C en su s d at a Ro bi nso n a nd W ea ve r ( 20 06 ); To m as ze w sk i, Ro bi ns on , W ea ve r, S tr yk er , a nd M ac Eac hr en (2 00 7) G eoT im e 20 07 C om m er ci al l aw e nf or ce m en t G VA sy st em ; s up po rt s m ult ip le d ev ic es (d es kt op , m ob ile , a nd w eb ‐ba se d) Sy nch ro no us di st rib ut ed , a nd as yn ch ro no us di st rib ut ed C rimina l in te lli ge nc e a na ly si s Pr ou lx e t a l. (2 00 7) ; E cc le s, K ap le r, H ar pe r, a nd W rig ht (2 00 8) ; U nc ha rte d (2 01 7) Sp ot fir e 20 07 Co m mer ci al g en er al‐ pu rp os e V A sy st em ; a va ila bl e a s d es kt op ‐ ba se d, w eb ‐ba se d, a nd c lou d‐ ba se d s ys te m ; s up po rt s m ul tip le de vi ce s A sy nch ro no us di st rib ut ed G en er al p ur po se ; r ep or te d ap pli ca tio ns in d om ai ns su ch as en er gy , f in anc ia l s er vi ce s, ma nuf ac tu ring , a nd tel ec omm un ic at io ns Ei ck e t a l. (2 00 7) ; V ie ga s et a l. (2 00 7) ; T IB C O (2 01 7) G eo A na ly tic s V is ua liz at io n ( G AV ) Fr am ew or k 20 08 Fr am ew or k a nd c la ss l ib ra ry f or ra pi d de ve lo pmen t o f w eb ‐ba se d G VA a pp lic at io ns A sy nch ro no us di st rib ut ed Ex plo ra tio n a nd a na ly si s o f st at is tic al in dic at or s, a na ly si s o f vo lu m et ric da ta , a nd a na ly si s o f flo od d at a, a m on g o the rs Je rn (2 00 8, 2 00 9, 2 01 0) ; H o a nd Je rn (2 01 3) ; L un db la d (2 01 3) G eo V iz T oo lki t 20 09 D es kto p‐ ba se d g ene ra l‐p ur po se G VA s ys te m Sy nch ro no us di st rib ut ed A na ly si s o f h ea lth ‐r el ate d d at a an d o f t er ro ris t a tt ac k d at a H ar di st y ( 20 09 ); H o e t a l. ( 20 12 ) RE N C I G eo A na ly tic s Fr am ew or k 20 11 Cy be r‐ in fr as tr uc tu re fo r d ev elo p-men t o f w eb ‐b as ed G VA s ys te m s; su pp or ts m ult ip le d ev ic es Sy nch ro no us di st rib ut ed a nd as yn ch ro no us di st rib ut ed H ur ric an e d amag e a ss es sm en t (C yb er Ey e) , emer genc y m an ag em en t a nd r es po ns e ( B ig B oa rd) H ea rd (2 01 1) ; H ea rd et a l. (2 01 3) ; K ije w sk i‐C or rea et a l. (2 01 4)
Sy st em na m e Yea r D es cr ip tio n C ol la bor at iv e sc ena rio A pp lic at io ns Re fer en ce s Tab le au 201 2 Co m mer ci al g en er al‐ pu rp os e V A sy st em ; s er ver v er si on o ff er s co lla bo ra tiv e f ea tu re s a nd inte gr at io n f or d es kt op ‐ba se d, w eb ‐ba se d, a nd m ob ile v er si on s; su cc es so r o f P ol ar is s ys te m A sy nch ro no us di st rib ut ed G en er al p ur po se . R ep or tin g ap pli ca tio ns in d om ai ns su ch as ba nk in g, c omm un ic at io ns , ed uc at io n, g ov er nmen t, in su ra nc e, a nd s po rt s H o e t a l. (2 01 2) ; E lia s, A uf au re , a nd B eze ria no s (2 01 3) ; T ab le au (2 01 7) PLO A D 20 13 En vi ro nm en ta l d ec is io n s up po rt sy st em ( ED SS ); a va ila bl e a s w eb ‐ba se d a nd c lou d‐ ba se d sy st em ; f oc us es o n t he m an ag e-m ent o f w ate rs he ds A sy nch ro no us di st rib ut ed W at er sh ed m an ag emen t Su n (2 013 ) Q Lik 20 13 Co m mer ci al g en er al‐ pu rp os e V A sy st em ; a va ila bl e a s d es kt op ‐ ba se d, w eb ‐ba se d, a nd c lou d‐ ba se d; s up po rt s m ult ip le d ev ic es A sy nch ro no us di st rib ut ed G en er al p ur po se . R ep or te d ap pli ca tio ns in d om ai ns su ch as he al th c ar e, f in an ci al s er vi ce s, en er gy a nd u til iti es , a nd l ife sc ienc es Q Li kT ec h (2 01 7) IB M W at so n A na ly tic s 20 14 Co m mer ci al g en er al‐ pu rp os e V A sy st em ; c lou d‐ ba se d A sy nch ro no us di st rib ut ed G en er al p ur po se . R ep or te d ap pli ca tio ns in d om ai ns su ch as ba nk in g, in su ra nc e, re ta il, tel ec omm un ic at io n, a nd ed uc at io n M ill er (2 01 4) ; I B M (2 01 7) SA S V is ua l A na ly tic s 201 5 Co m mer ci al g en er al‐ pu rp os e V A sy st em ; o ff er s o n‐ pr em is es a nd clou d‐ ba se d depl oy men ts ; su pp or ts m ult ip le d ev ic es A sy nch ro no us di st rib ut ed G en er al p ur po se . R ep or te d ap pli ca tio ns in d om ai ns su ch as ba nk in g, c omm un ic at io ns , de fe ns e a nd s ec ur ity , h ea lth ca re , a nd h ig h‐ te ch ma nuf ac tu ring SA S (2 01 5, 2 01 7) Th e “ Ye ar ” c ol um n s ho w s t he y ea r o f t he fi rs t r ef er en ce to a co lla bo ra tiv e f ea tu re in th e s ys te m , if k no w n; th e “ D es cr ip tio n” a nd “A pp lic at io ns ” c ol um ns a re b as ed o n a st ud y o f t he li te ra -tu re r ef er en ci ng t he s ys te m , a nd f re el y a va ila bl e i nf or m at io n f ro m t he s of tw ar e p ub lis he r; a nd t he “ C ol la bo ra tiv e s ce na rio ” c ol um n f ol lo w s t he w id el y a do pt ed c at eg or iz at io n p ro po se d by J oh an se n ( 19 88 ), w ho d is tin gu is he s b et w ee n s yn ch ro no us a nd a sy nc hr on ou s c ol la bo ra tio n, a nd d is tr ib ut ed a nd c o‐ lo cate d w or kp la ce s. Table 2 ( Con tinued)
and space, significantly increasing the potential scalability of the collaborative effort (Heer, van Ham, Carpendale, Weaver, & Isenberg, 2008); additionally, the study by Benbunan‐Fich, Hiltz, and Turoff (2003) found that asynchro-nous collaboration resulted in higher‐quality outcomes because participants have time to generate and reflect on new ideas, and can contribute regardless of their location. Synchronous and co‐located (8% of the systems) requires specific hardware for parallel input from multiple sources and parallel output to a potentially diverse audience to enable it, requiring a more demanding design of the interface. Asynchronous and co‐located (8% of the systems) means effectively sharing the same input and output devices, which has become rare with falling hardware costs. Lastly, distributed but synchronous (23% of the systems) is more common, but requires special coordination across different locations and potentially time zones.
These collaborative scenarios are not mutually exclusive, and a combination is called a hybrid collaborative sce-nario. As shown in Figure 2, a mixed‐presence scenario has co‐located and distributed users (Marrinan et al., 2017), and a multi‐synchronous scenario features synchronous and asynchronous interactions (Preguiça, Martins, Domingos, & Duarte, 2005). Isenberg et al. (2011) claims the need to expand the research in hybrid collaborative scenarios, which is supported by our finding that only 23% of the identified systems support some hybrid collaborative scenario.
The literature argues that the support for scenarios of time‐critical and long‐term analysis are of key importance for CGVA (Andrienko et al., 2007; Isenberg et al., 2011; Keim et al., 2010; Thomas & Cook, 2005). These analysis scenarios are characterized by the duration of the analysis effort. In a time‐critical scenario, the analysis has to be completed as rapidly as possible to minimize undesirable consequences. Examples are analysis in response to natu-ral disasters or terrorist attacks. Among the identified systems, only the RENCI GeoAnalytics Framework (Heard, 2011) is designed to support collaborative time‐critical analysis of emergency situations. In a long‐term scenario, the analysis extends over a longer time span and usually aims to generate understanding and/or enable strategic deci-sions. Examples are analysis of climate change and species conservation. None of the identified systems is designed to support a long‐term scenario.
F I G U R E 2 Collaborative analysis in GVA can occur in four scenarios defined by space and time. The figure
shows those scenarios and the systems that support them. Additionally, hybrid collaborative scenarios (e.g., mixed presence and multi‐synchronous) are shown
Co-located Distributed su on or hc ny S su on or hc nys A
ORACLE BI Visual Analycs SAP BusinessObjects* Spoire
GeoAnalycs Visualizaon (GAV) Tableau
PLOAD Qlik
IBM Watson Analycs SAS Visual Analycs SAP BusinessObjects* GeoTime GeoViz Toolkit ReVise
1
3
1
11
Space
e
miT
* SAP BusinessObjects appears twice, it was done to avoid confusion by placing it in a diagonal between synchronous co-located and asynchronous distributed.
** Collaborave scenarios are not mutually exclusive, for this reason the column No do not sum up to 13 and % to 100%. Mixed-presence Mul -s ynch ro nous Collabora ve scenario** No % Synchronous co-located 1 8% Synchronous distributed 3 23% Asynchronous co-located 1 8% Asynchronous distributed 11 85%
Hybrid collabora ve scenario No %
Mixed-presence 0 0%
Mul-synchronous 2 15%
Synchronous co-located and
asynchronous distributed 1 8%
No = Number of systems that support the scenario % = Percentage of systems that support the scenario
(out of 13 systems) Rency GeoAnalycs
Our analysis shows that GVA environments are increasingly using cloud‐based platforms, as shown in Figure 3. This is a general trend in analytical systems (Vesset, 2016), and offers two advantages: flexible and scalable storage capacity and processing power to work with large and complex datasets, enabling users to work from thin clients; and the distributed access to the system enabled by the internet, which improves the potential for multidisciplinary and cross‐domain collaboration among geographically separated participants.
Finally, most of the identified systems support multiple device types such as PCs, smartphones, tablets, touch tables, or large screens (explicit claims were identified for 62% of the systems). This is enabled by the usage of web‐ based interfaces and facilitates reaching a broader audience, which is further promoted by most systems support-ing asynchronous distributed collaboration and increassupport-ing usage of cloud‐based deployment. This combination of technologies removes the constraints of time and space to participate in analysis efforts, and eliminates the need for specialized hardware to access the system. The concept of contributing irrespective of location, device, or time is called ubiquitous analytics (Elmqvist & Irani, 2013). This convergence of technologies, which dramatically improves the potential for effective collaboration in the analysis of geographic information, was predicted almost two decades ago by MacEachren (2000).
3.2 | Collaborative techniques
We identified six collaborative techniques: annotation, discussion board, instant messaging, interaction history, snapshot, and storytelling. Table 3 offers a summary of the advantages and limitations of each technique. Among these techniques, snapshot, storytelling, and annotation are the most popular ones, implemented by 85, 62, and 54% of the systems, respectively (see Table 4). Further, they are the techniques that co‐occurred more often (see Table 5a). The combination of these three techniques offers a flexible working environment that allows analysts a seamless combination of independent and collaborative analysis, and self‐explanatory results that can be immedi-ately communicated. Finally, we created a cross‐tabulation of collaborative techniques and scenarios through the systems supporting them (see Figure 5b later). Noteworthy is the absence of the annotation technique in the system
F I G U R E 3 The usage of cloud technology to deploy CGVA systems is increasing, which improves the scalability
and distributed access to the system
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
Desk
to
p-ba
se
d
We
b-ba
se
d
Clou
d-ba
se
d
ReVise Ge oTim e GA V Ge oViz Toolki t RE NC I Ge oAnaly c s Tablea u IBM Wa ts on Anal y cs PLOA D QL ik Spo ir e SAS Visual Analy cs Im prov ed di st ribut ed access Im prov ed scal abilit ysupporting a synchronous co‐located scenario, because this can certainly help co‐located participants to support claims during a discussion session. Additionally, snapshot is the only technique that occurs in all four scenarios, which provides evidence of its flexibility.
3.2.1 | Annotation
Annotation means any piece of information in the form of text or graphic, attached to an information product such as
a data table, illustration, or map. An annotation may be a mere overlay on a visual product, but it may also be a data‐ aware artefact carrying semantics that link the annotation with the underlying data (Heer & Shneiderman, 2012; Ren, Brehmer, Lee, Höllerer, & Choe, 2017). Annotations can be made on the aggregate level of the information prod-uct, or on individual features comprising them (Ren et al., 2017). In (geo)visualization, annotation facilitates access to and recall of contributions (i.e., external memory), document ideas in private and public discussion spaces, and elicit information from all participants in a collaborative effort (Heer & Shneiderman, 2012; Hopfer & MacEachren, 2007).
TA B L E 3 Advantages and limitations of the identified collaborative techniques
Technique Advantages Limitations
Annotation
·
Enables analysts to point at, describe, and bound features of interest in the data products·
Can carry semantics that link the annotation with the underlying data·
Lack of guidelines to regulate its use may lead to an overload of irrelevant contributionsDiscussion board
·
Enables topic‐centered discussion among geographically distributed analysts·
Topics are organized in threads·
Synthesized discussion results is not trivial Instant messaging·
Enables discussion among geographicallydistributed analysts
·
Private discussions may lead to lack of awareness of others’ work and to fragmenta-tion of the known informafragmenta-tion·
Discussion board is more flexible and better organizedInteraction history
·
Documents the analysis as a continuous process automatically·
The interaction logs can be stored and accessed based on different models·
Allows the analysis process to be reviewed and extended·
An interaction history may require editing before it can be disseminated·
Snapshot offers an alternative to document the analysis process as discrete states and is deemed sufficient in most use casesSnapshot
·
Allows discrete states of the analysis process to be stored on demand·
Stored states can be reconstructed for further analysis·
Can be applied to independent visual products or the whole analytical environment·
Unlike interaction history, snapshot cannot reconstruct the interactions that led to the stored statesStorytelling
·
Organized in chapters·
Supports a flexible analysis process by allowing the story to be updated·
Specific focus on communication of analytical results·
Effective, engaging, and easy to understand for specialists and laypersons·
Doesn't incorporate identification of individual's contributionsAnnotation has three main functions (Heer & Shneiderman, 2012; Hopfer & MacEachren, 2007): (a) to highlight a feature of interest in a visual product (e.g., a potentially suitable location for facilities); (b) to provide information on the feature of interest (e.g., describing building status during post‐disaster damage assessment); and (c) to act as a bounding object (e.g., delineating areas affected by a natural disaster). Figure 4 shows examples of annotation in the context of pest management.
Annotation is implemented by 54% of the systems, and applies to every loop of the analysis process. During the exploration loop, it enables analysts to highlight, describe, and communicate findings. Later, in the verification loop, these findings are interpreted in the problem's domain and constitute insights that may lead to hypothesis genera-tion or to identifying evidence for existing hypotheses. Insights can be documented and communicated using anno-tation. In both loops, annotation creates awareness of the findings and insights, and constitutes documentation for the acceptance or rejection of a hypothesis in the knowledge generation loop. Two aspects for efficient use of anno-tation are: first, guidelines to moderate the usage of annoanno-tations, with the lack of them possibly creating an overload of irrelevant contributions (Hopfer & MacEachren, 2007); second, functionality to track existing annotations, create
TA B L E 4 List of the identified collaborative techniques and the GVA systems implementing them
Annotation Discussion board Instant messaging Interaction history Snapshot Storytelling
Total of tech-niques/ system ORACLE BI Visual Analytics ✓ ✓ 2 SAP BusinessObjects ✓ ✓ ✓ 3 ReVise ✓ ✓ ✓ 3 GeoTime ✓ ✓ ✓ 3 Spotfire ✓ ✓ ✓ 3 GeoAnalytics Visualization (GAV) Framework ✓ ✓ 2 GeoViz Toolkit ✓ ✓ 2 RENCI GeoAnalytics Framework ✓ ✓ 2 Tableau ✓ ✓ ✓ 3 PLOAD ✓ 1 QLik ✓ ✓ 2 IBM Watson Analytics ✓ ✓ ✓ 3 SAS Visual Analytics ✓ ✓ ✓ 3 Total of systems/ method 7 3 2 1 11 8 Percentage of systems with method 54% 23% 15% 8% 85% 62%
T A B LE 5 (a) C o‐ oc cu rr en ce o f c ol la bo ra tiv e t ec hn iq ue s i n t he i de nt ifi ed s ys te m s; v al ue s i n b ol d s ho w t he t w o h ig he st c o‐ oc cu rr en ce s o f t ec hn iq ue s. ( b) N um be r o f sy st em s t ha t i m pl em en t a te chn iq ue an d s up por t a c ol lab or at iv e s ce nar io (a ) (b ) A nn ot at io n D is cu ss io n bo ar d Ins ta nt m ess ag ing In te ra cti on hi sto ry Sna ps ho t St or yt ellin g Sy nc hr ono us co ‐loc at ed Sy nc hr ono us di str ibu ted A sy nc hr ono us co ‐loc at ed A sy nc hr ono us di str ibu ted A nn ot at io n – – – – – – 0 2 1 6 D is cu ss io n bo ard 1 – – – – – 1 0 0 3 In st an t m ess ag in g 1 0 – – – – 0 2 0 1 In ter ac tio n hi sto ry 1 0 0 – – – 0 0 1 0 Sna ps ho t 5 3 1 1 – – 1 2 1 9 St or yt elli ng 3 2 0 0 8 – 1 1 0 8
links between annotations to understand their relationships, and synthesize them (Cai & Yu, 2009; Wu, Convertino, Ganoe, Carroll, & Zhang, 2013).
3.2.2 | Discussion board
A discussion board (also known as a discussion forum) enables users to exchange text messages on a chosen topic. Since users are allowed to reply directly to any message, the communication is not necessarily linear, but follows a hierarchical structure, in which each branch is called a thread (Weinberger & Fischer, 2006; Yusoff & Salim, 2015).
Figure 5 shows an example of a discussion board in a GVA environment.3
A discussion board can be used in all the loops of the analysis process. It provides a mechanism to discuss ideas, generate hypotheses, share findings, reach agreements, and plan further actions. Analysts can create threads to dis-cuss findings (exploration loop). These threads document the arguments to understand the findings in the context of the analysis domain, which constitutes insights (verification loop). These insights can lead to generated hypotheses or identified evidence, and analysts can create threads to organize and document them. Once enough evidence is available, the analysts can use the content of the threads as input to draw conclusions (knowledge generation loop).
In comparison with instant messaging, a discussion board enables analysts to engage in different discussions around the same data view without mixing the topics, because each topic has its own thread. As a benefit of offering a better organization for discussions, these threads make the message board more suitable to support larger groups of participants. Additionally, the discussions are public, which ensures transparency among all the analysts. While both techniques are based on the idea of message exchange, these differences make the message board slightly more popular, with 23% of the identified systems implementing it, against 15% implementing instant messaging.
Despite threads improving the organization of the discussions, synthesizing them can be a cumbersome task due to unstructured contributions, ambiguity, and unclear references. This issue can be addressed by adopting a formal argumentation model, and examples include Luther, Counts, Stecher, Hoff, and Johns (2009), Willett, Heer, Hellerstein, and Agrawala (2011), Rinner (2001), and Rinner, Keßler, and Andrulis (2008). The latter is particularly
F I G U R E 4 Examples of the usage of annotation to point to a feature of interest (i.e., location with highest
measurements), to describe a feature of interest (i.e., damage in olive fruit), and as a bounding object (i.e., area that requires control measurements)
Monitoring locaon with
highest measurements
through out the year
Consider applicaon of
control measurements
relevant, as it describes “Argumentation maps” or “Argumaps” combining the strengths of argumentation modeling and detailed geographic location to support any argumentation process that has a spatial component.
3.2.3 | Instant messaging
Instant messaging enables users on a network to exchange text messages (Chatterjee, Abhichandani, Haiqing, TuIu,
& Jongbok, 2005). It was originally designed as a one‐on‐one synchronous communication method, but nowadays it also serves for discussions among more than two participants and for asynchronous communication. It was popular-ized originally in the late 1990s by systems such as America Online's Instant Messenger (AIM), Microsoft Messenger, Yahoo! Messenger, and more recently by Facebook Messenger and WhatsApp (Desjardins, 2016; Petronzio, 2012). In addition to text‐based communication, current implementations allow the inclusion of hypertext, multimedia ele-ments, and file exchange. In the context of collaborative analysis, instant messaging enables analysts to work to-gether to solve analytic problems regardless of their location (Hardisty, 2009).
Like discussion boards, instant messaging provides a mechanism to discuss ideas, generate hypotheses, share findings, reach agreements, and plan further actions, during all analysis phases. However, the private and directed nature of instant messaging communications limits their usefulness in a collaborative setting for two reasons: first, lack of awareness of others’ work that may lead to duplicated efforts; and second, fragmentation of the known infor-mation that may make it difficult to share a common ground for the analysis. Both awareness and common ground are key elements for effective collaborative analysis (Heer & Agrawala, 2008). From the point of view of DC, a key element is the external representation of information/knowledge and its propagation (Susi & Ziemke, 2001). While the exchanged messages constitute externalization and propagation of information/knowledge, these are limited to the participants of the instant messaging session. Therefore, messages don't propagate (at least directly) to the other analysts, limiting the cognitive activity of the system as a whole.
F I G U R E 5 Spotfire allows us to create and access discussion boards linked to a visual product or from a
Figure 6 shows the interface of the GeoViz Toolkit and its instant messaging component called GeoJabber.4
An interesting feature of GeoJabber is that analysts can share the current status of the visual interface using the technique of snapshot to support claims during discussion sessions (Hardisty, 2009).
Despite instant messaging being a widely used communication technique, our literature review shows that its usage in CGVA platforms is quite limited, and only 15% of the identified systems implement it. Two reasons explaining its limited use are: first, awareness and common ground are of key importance for collaborative work, and instant messaging may disrupt them; and second, message boards offer equivalent functionality but are more flexible.
3.2.4 | Interaction history
Interaction history provides analysts with the capacity to save, review, and reuse analytical work. To do so, this
tech-nique creates logs of the user's actions and/or state changes in a GVA environment during analysis sessions (Heer, Mackinlay, Stolte, & Agrawala, 2008). These logs can be organized based on different models such as stack, linear, and branching (Heer, Mackinlay et al., 2008; Lundblad, 2013). The stack model is the simplest and enables analysts to undo and redo actions/states. The linear model stores the actions/states in the order of occurrence and enables analysts to transverse the analysis as a linear continuum. The branching model stores the actions/states in a tree‐like structure, which allows documenting multiple analysis paths.
We only identified one implementation of interaction history, the re‐visualization technique in ReVise (Robinson & Weaver, 2006). This allows saving and revisiting session logs, and offers options called “jump in” and “bread-crumbs.” Jump in allows us to resume an analysis session at any given moment and extend it. Breadcrumbs allows us
to create indicators of key moments in the analysis session, and attach annotations based on text or audio to describe why the analyst considers them of importance.
Interaction history supports all the loops of the analysis process by allowing the analysts’ interaction with the system to be documented. Among the identified techniques, this is the only one that automatically and unobtru-sively documents the analysis as a continuous process, and further allows us to review and extend it. The general role of this technique is to enable analysts to document and review the analysis process that led to findings, insights, hypotheses, evidence, and conclusions. Additionally, it can help in better understanding individual and collaborative strategies during data analysis (Heer, Mackinlay et al., 2008).
This is the least popular among the identified techniques, with only one implementation (8% of the systems). A likely reason is that the conceptually similar—but technically less complex—snapshot technique is implemented by most systems and deemed sufficient for most use cases. They both allow us to document the evolution of the analysis process, but they differ in the level of detail. While interaction history documents the analysis as a continuous pro-cess and allows us to review all interactions and state changes, snapshot only captures discrete states. Another dif-ference is that interaction history automatically documents the analysis process, while snapshot documents states on demand. Another factor limiting its utility is the need to edit the history of the analysis and exploration process (e.g., select only relevant parts and add narrative) before it can be disseminated (Lundblad, 2013).
3.2.5 | Snapshot
The snapshot technique captures the state of a visual product at a given moment, which can be used to document and share findings and insights (e.g., patterns and outliers) for further analysis or communication (Lundblad, 2013). A simple and common approach is to capture it as a static image (e.g., ReVise; Robinson & Weaver, 2006). However, there are more advanced approaches that store different parameters to allow reconstruction of the captured state (e.g., GeoJabber; Hardisty, 2009), which allows further exploration and analysis from the stored state. Furthermore, this concept can be extended to capture the state of the whole analytical environment composed of several visual products, as implemented in the GeoAnalytics Visualization (GAV) Framework (Ho & Jern, 2013). Figure 7 shows a schematic view of the mechanism to capture and restore snapshots in the GAV Framework.
F I G U R E 7 The GAV Framework is capable of recording the state of all components in the analytical
environment as XML; later this file can be used to reconstruct the state of the environment for further analysis. Illustration based on Ho, Lundblad, Astrom, and Jern (2012)
GeoVisual Analyc Environment Map component Scaer plot component
…
Parallel coordinate plot component Createsnapshot snapshotRestore
Snapshots.XML
Current status Stored status
Get components’
status Set components’ status Get environment’s status Set environment’s status
The snapshot technique is commonly used for asynchronous collaboration, allowing analysts to reconstruct saved states on demand. However, GeoJabber implements a synchronous version to support claims during analysis sessions supported by instant messaging (Hardisty, 2009). To do so, GeoJabber includes a mechanism to capture, encode, and transfer the snapshot as a special type of message, which modifies the receiver's viewer to reconstruct the sender's view.
Like interaction history, snapshot can be used in all the loops of the analysis process, and enables analysts to document the evolution of the analysis process. Unlike interaction history, snapshot only captures specific states of the analysis process, and usually is manually triggered.
Snapshot is the most popular among the identified techniques (85% of the systems), and is commonly used in combination with storytelling. In our review, all eight systems that implement storytelling also implement snapshot, which represents the highest co‐occurrence of techniques. Additionally, the second highest co‐ occurrence is between snapshot and annotation (five systems). This combination allows us to describe the analysis process through a “story” (Jern, 2010), and support the claims on it by using snapshots to document findings, and annotation to highlight and describe specific aspects (Viegas, Wattenberg, van Ham, Kriss, & McKeon, 2007).
3.2.6 | Storytelling
Storytelling is a comprehensive approach combining methods to tell a story about data exploration and the
analy-sis process that led to certain findings or conclusions through interactive visualization (Jern, 2010). A story may be organized in chapters, and include descriptions, multimedia elements, annotations, and snapshots, all facilitating a reader's understanding of the original analytical process (Lundblad, 2013).
Storytelling can be seen as a communication technique. However, by allowing the reader to interact with the snapshots in the story, s/he can explore further, create new snapshots, and modify the story with new findings and insights. In this case, storytelling is not only a communication technique, but also a method for collaborative knowl-edge building (Lundblad, 2013). Figure 8 shows the storytelling technique in an application (http://mitweb.itn.liu.se/ GAV/dashboard/#story) developed with the GAV Framework.
Storytelling (62% of the systems) is applicable to every loop of the analysis process, and is commonly used in combination with snapshot and annotation techniques. This combination provides a working space in which analysts describe the analysis process through a story (Jern, 2010), and document relevant observations using snapshot and annotation (Viegas et al., 2007). Given that the story can be updated with new findings, insights, hypotheses, and ev-idence as needed, storytelling supports a more flexible analysis process. This is not possible with instant messaging or message board, because they are based on the idea of appending contributions and not modifying them.
The convenience of this combination of techniques can be explained from the point of view of DC. In this context, the cognitive artefacts are highly important because they represent externally an individual's internal representa-tion of informarepresenta-tion/knowledge, and thereby enable cognirepresenta-tion across individuals and time (Hollan et al., 2000; Susi & Ziemke, 2001). The combination of storytelling, snapshot, and annotation constitutes an effective media for external-ization and propagation because it combines the flexibility of snapshots to externalize the context with the attention to specific details and relationships of annotations, while exposing the entire reasoning process through storytelling. Further, unlike instant messaging, it is publicly available, which ensures propagation of the externalized representa-tions. Lastly, it is succinct, because the story evolves to keep only the relevant information, unlike a discussion board that appends all the contributions, or interaction history that records all interactions and state changes.
Among the identified techniques, storytelling is the only one with specific focus on effective communication of analytical results. It offers a flexible working space for independent and collaborative analysis, where results (i.e., stories) can be immediately communicated to a broader audience. Authors agree that results presented with storytelling are more effective, engaging, and easy to understand for specialists and laypersons (Grainger, Mao, & Buytaert, 2016; Stodder, 2015; Sun, Wu, Liang, & Liu, 2013).
3.3 | Synthesis of results
The most common collaborative scenario is asynchronous distributed, because analysts are not constrained by time or space from participating in the analysis. This increases the potential for scalability of the collaborative effort, and improves the quality of analytical results. We also identified limited support for hybrid collaborative scenarios, and time‐critical and long‐term analysis scenarios. Regarding the technological platform, the usage of cloud technology is increasing, which improves the scalability and accessibility of the systems. In terms of supported devices, the usage of web‐based interfaces enables the majority of the systems to support multiple devices such as PCs, smartphones, tablets, touch tables, and large screens.
The most commonly supported collaborative techniques are snapshot, storytelling, and annotation, which also co‐occur often. This combination of techniques offers a flexible setup to build knowledge through an iterative process. Storytelling offers a working space in which analysts describe the analysis process through a story, and document rele-vant observations using snapshot and annotation. Snapshot captures the context of an observation, and annotation high-lights and describes specific aspects of it. During the analysis process, the story is updated as new findings and insights are produced, and once it is concluded, the story can be used immediately to communicate results to a broader audience.
Despite the fact that the identified techniques support the three loops of the analysis process, we didn't identify any mechanisms to aid in the synthesis of analytic results, which is important to support the knowledge generation loop, or to summarize the level of agreement about the evidence and conclusions, which would provide certainty when results are communicated.
4 | RESEARCH CHALLENGES
Following from our systematic review, we argue that the three most pressing challenges to support collaborative work in GVA systems are: hybrid collaborative scenarios; cross‐device collaboration; and support for time‐critical and long‐term analysis.
4.1 | Challenge 1: Hybrid collaborative scenarios
Collaborative systems are commonly characterized by the time and space in which collaboration takes place. This characterization of systems defines four collaborative scenarios (see Figure 2). Any combination of these scenarios is a hybrid collaborative scenario [e.g., mixed presence supporting co‐located and distributed participants (Marrinan et al., 2017, 2016) or multi‐synchronous supporting synchronous and asynchronous contributions (Preguiça et al., 2005)]. Already in 2011, Isenberg et al. (2011) claimed the need to further research hybrid scenarios in information visualization and raised the expectation of seeing more systems supporting them in the following years. This review article shows that it remains a challenge in CGVA, with only 23% of the identified systems supporting hybrid col-laborative scenarios.
To support collaborative work effectively and efficiently in CGVA, it is necessary to consider that a typical analysis effort comprises many different tasks, each of which may benefit or even require more than one collabo-rative scenario. Therefore, instead of forcing analysts to work in a specific collabocollabo-rative scenario, CGVA systems should enable them to move seamlessly between scenarios. For example, in the domain of emergency management, co‐located and distributed analysts need to collaborate in real time during the emergency situation, while during the relief stage asynchronous collaboration may be more suitable. In this example, the analysts benefit from a hy-brid scenario related to the location (i.e., mixed presence) and another one related to the time of collaboration (i.e., multi‐synchronous).
To address this challenge, future research needs to investigate the suitability of collaborative scenarios for spe-cific types of task, evaluate the advantages and disadvantages of hybrid scenarios which may depend on the appli-cation domain, and design mechanisms that allow analysts to move seamlessly from one collaborative scenario to another while keeping all the analysis results available and ensuring awareness of others’ work. Additionally, special attention is required for scenarios in which analysts may work offline, which may require temporal local storage and a versioning system, so that the client's devices can synchronize when connection is available.
4.2 | Challenge 2: Cross‐device collaboration
The support for multiple types of device has the potential to provide analysts with a more flexible analysis workflow, engage a more diverse audience, and facilitate collaboration (Elmqvist, 2014); 62% of the identified systems explic-itly claim to support multiple types of device (see Table 2). However, there is little evidence of those systems taking advantage of the unique characteristics of each type of device. For example, most of the identified systems support the use of smartphones, but only as viewers. In contrast, Big Board (Heard, Thakur, Losego, & Galluppi, 2013) allows the use of integrated sensors on smartphones to capture information (e.g., photos, videos, and sounds) and create geo‐located annotations to share it.
The support for cross‐device collaboration also has the potential to improve multidisciplinary and cross‐domain analysis by enabling actors from diverse backgrounds (e.g., scientists, domain experts, and laypersons or citizen sci-entists) to participate without requiring specialized devices or specific hardware (Badam, Fisher, & Elmqvist, 2015; Elmqvist & Irani, 2013). To provide effective support for cross‐device collaboration, the user interface needs to take advantage of the unique characteristics of diverse types of device. For example, in the analysis of species distribu-tion, an in‐office analyst may benefit from the use of desktop workstations to identify features of interest such as patterns and outliers, and to develop a hypothesis, while an in‐field analyst may benefit from the use of smartphones to check for the status of the species’ population in his/her surroundings, and to capture information that may act as evidence. Especially in the growing field of citizen science, it is a crucial element of any project to let participants with diverse skills, capabilities, interests, and hardware collaborate.
To address this challenge, research is required to evaluate the capacity and limitations of each type of device in the context of CGVA, analyze the activities that each type of device may support, design mechanisms that allow participants to move seamlessly from one device to another while all contributions remain available, and develop an
infrastructure that ensures responsiveness regardless of the device in use. Additionally, research is needed to evalu-ate the suitability of the devices in relation to the diverse collaborative scenarios.
4.3 | Challenge 3: Time‐critical and long‐term analysis
The support for time‐critical and long‐term analysis scenarios has been argued to be of key importance for CGVA (Andrienko et al., 2007; Keim et al., 2010; Thomas & Cook, 2005). These scenarios are defined by the duration of the analysis.
Only one of the identified systems is designed to support time‐critical analysis, and none long‐term analysis. While it can be argued that any system can be used in both scenarios, the lack of specialized functionality hampers the flow of the analysis process. For example, in an emergency situation (a time‐critical scenario), analysts need specialized tools to deal with rapidly changing analysis conditions (i.e., real‐time data updates, dynamic planning and coordination, and awareness of others’ work), and tools to negotiate and reach consensus in an agile manner (Andrienko et al., 2007). These are not expected characteristics from a general‐purpose CGVA system, and poten-tially not the ones required for long‐term analysis.
Given the abundance of applications that can benefit from these analysis scenarios, and the need for specialized functionality to effectively support them, we recommend further research to design and evaluate systems for time‐ critical analysis that prevent conflicting interaction due to the concurrent access to resources during the analysis sessions, ensure awareness of all the participants regarding the progress of the analysis, and facilitate timely com-munication of results. Additionally, research into the design of systems for long‐term analysis that properly sum-marize progress—such that participants can easily catch up with the advances from one session to another—allow participants to work in parallel with multiple analysis projects and potentially with multiple working hypotheses in each one, and facilitate the communication of partial and final results.
5 | CONCLUSIONS
Three developments show that CGVA environments are aiming to reach a broader audience: First, the most common collaborative scenario is asynchronous distributed, which promotes participation by removing the constraints on time and location to contribute; second, the increasing use of cloud technology, which improves distributed access to the system; and third, the support for multiple devices, which eliminates the need for specialized hardware.
The most common collaborative techniques are snapshot, storytelling, and annotation, which are supported by almost all systems, and complement each other. The combination of these techniques offers a flexible working envi-ronment that allows analysts a combination of independent and collaborative analysis, and self‐explanatory results that can be communicated immediately. The other techniques are uncommon due to either lack of flexibility, diffi-culty of integrating into the system interface, or implementation requirements.
The features offered by the reviewed CGVA environments support the whole process of data analysis, including identification of features of interest, generation of hypotheses, provision of evidence, and communication of ana-lytic results. However, features are missing to aid the synthesis of anaana-lytic results, which is important to support the knowledge generation loop, and to summarize the level of agreement about the evidence and conclusions, which would provide certainty when results are communicated.
We can conclude that CGVA is moving toward effective support of multidisciplinary and cross‐domain collabora-tive analysis. However, for this potential to materialize, researchers should address three pressing challenges: hybrid collaborative scenarios, cross‐device collaboration, and support for time‐critical and long‐term analysis.
NOTES
1A Supporting Information document is available with the search and selection procedure on each database.
2The total number of matches for the query was 740, but Google Search detected that 56 were the most relevant results, and the others were very similar. This result is not reproducible, because Google Search results are based on many unknown variables, including individual user search history.
3Screenshot created using the Sales and Marketing example included in Spotfire. 4Screenshot created using the Health example included in the software.
ORCID
Gustavo Adolfo García‐Chapeton http://orcid.org/0000-0001-6709-4783
Frank Olaf Ostermann http://orcid.org/0000-0002-9317-8291
REFERENCES
Andrienko, G., Andrienko, N., Jankowski, P., Keim, D., Kraak, M., MacEachren, A., & Wrobel, S. (2007). Geovisual analytics for spatial decision support: Setting the research agenda. International Journal of Geographical Information Science,
21(8), 839–857.
Andrienko, N., & Andrienko, G. (2013). Visual analytics of movement: An overview of methods, tools and procedures.
Information Visualization, 12(1), 3–24.
Arimoto, M. M., & Barbosa, E. F. (2012). A systematic review of methods for developing open educational resources. In
Proceedings of the 20th International Conference on Computers in Education. Singapore.
Badam, S. K., Fisher, E., & Elmqvist, N. (2015). Munin: A peer‐to‐peer middleware for ubiquitous analytics and visualization spaces. IEEE Transactions on Visualization & Computer Graphics, 21(2), 215–228.
Benbunan‐Fich, R., Hiltz, S. R., & Turoff, M. (2003). A comparative content analysis of face‐to‐face vs. asynchronous group decision making. Decision Support Systems, 34(4), 457–469.
Cai, G., & Yu, B. (2009). Spatial annotation technology for public deliberation. Transactions in GIS, 13, 123–146.
Chatterjee, S., Abhichandani, T., Haiqing, L., TuIu, B., & Jongbok, B. (2005). Instant messaging and presence technologies for college campuses. IEEE Network, 19(3), 4–13.
Çöltekin, A., Bleisch, S., Andrienko, G., & Dykes, J. (2017). Persistent challenges in geovisualization: A community perspec-tive. International Journal of Cartography, 3, 115–139.
De Amicis, R., Conti, G., Piffer, S., & Simöes, B. (2009). Geospatial visual analytics. Dordrecht, the Netherlands: Springer. Desjardins, J. (2016). The evolution of instant messaging. Retrieved from http://www.visualcapitalist.com/
evolution-instant-messaging/
Eccles, R., Kapler, T., Harper, R., & Wright, W. (2008). Stories in geotime. Information Visualization, 7(1), 3–17.
Eick, S. G., Eick, M. A., Fugitt, J., Horst, B., Khailo, M., & Lankenau, R. A. (2007). Thin client visualization. In Proceedings of the
Second IEEE Symposium on Visual Analytics Science and Technology. Sacramento, CA: IEEE.
Elias, M., Aufaure, M.‐A., & Bezerianos, A. (2013). Storytelling in visual analytics tools for business intelligence. In P. Kotzé, G. Marsden, G. Lindgaard, J. Wesson, & M. Winckler (Eds.), Human–Computer Interaction – INTERACT 2013 (Lecture Notes in Computer Science) (Vol. 8119, pp. 280–297). Berlin, Germany: Springer.
Elmqvist, N. (2014). Visualization reloaded: Redefining the scientific agenda for visualization research. In Proceedings of the
Human–Computer Interactions Conference (pp. 132–137). Seoul, South Korea.
Elmqvist, N., & Irani, P. (2013). Ubiquitous analytics: Interacting with big data anywhere, anytime. Computer, 46(4), 86–89. Grainger, S., Mao, F., & Buytaert, W. (2016). Environmental data visualisation for non‐scientific contexts: Literature review
and design framework. Environmental Modelling & Software, 85, 299–318.
Haklay, M. (2013). Citizen science and volunteered geographic information: Overview and typology of participation. In D. Sui, S. Elwood, & M. Goodchild (Eds.), Crowdsourcing geographic knowledge (pp. 105–122). Dordrecht, the Netherlands: Springer.
Hardisty, F. (2009). Geojabber: Enabling geo‐collaborative visual analysis. Cartography & Geographic Information Science,
36(3), 267–280.
Heard, J. (2011). Geoanalytics (Report No. TR‐11‐03). Chapel Hill, NC: Renaissance Computing Institute.
Heard, J., Thakur, S., Losego, J., & Galluppi, K. (2013). Big board: Teleconferencing over maps for shared situational aware-ness. Computer Supported Cooperative Work 23(1), 51–74.
