2009-Social computing-First steps to netviz nirvana

First Steps to NetViz Nirvana: Evaluating Social
Network Analysis with NodeXL
Elizabeth M. Bonsignore∗§ , Cody Dunne ∗† , Dana Rotman∗§ , Marc Smith¶ ,
Tony Capone , Derek L. Hansen∗‡ , & Ben Shneiderman∗†
∗University of Maryland, College Park, MD 20742
§ Center for Advanced Study of Communities & College of Information Studies {drotman, ebonsign}@umd.edu
† Human-Computer Interaction Lab & Department of Computer Science {ben, cdunne}@cs.umd.edu
¶ Telligent Systems Corporation marc.smith@telligent.com

Microsoft Corporation v-tonyc@microsoft.com
‡ dlhansen@umd.edu 301-405-7622 2118C Hornbake Library

Abstract—Social Network Analysis (SNA) has evolved as a individual roles they might not otherwise see.
popular, standard method for modeling meaningful, often hidden This paper relates the process and early results of 2 user
structural relationships in communities. Existing SNA tools often studies, in which graduate students of information science
involve extensive pre-processing or intensive programming skills
that can challenge practitioners and students alike. NodeXL, an (IS) and computer science (CS) learned SNA concepts and
open-source template for Microsoft Excel, integrates a library of techniques while using NodeXL. Our aim is to focus on the
common network metrics and graph layout algorithms within the unique features that made NodeXL learnable and usable. First,
familiar spreadsheet format, offering a potentially low-barrier- we provide an overview of the NodeXL tool. We also describe
to-entry framework for teaching and learning SNA. We present an emergent research method called Multi-dimensional In-
the preliminary findings of 2 user studies of 21 graduate students
who engaged in SNA using NodeXL. The majority of students, depth Long-term Case studies (MILCs), an ethnography-based
while information professionals, had little technical background approach that seems well-suited to enabling more effective
or experience with SNA techniques. Six of the participants had evaluations of complex visual analytics tools ([3-7]). Next, we
more technical backgrounds and were chosen specifically for discuss our methodology and present visualizations produced
their experience with graph drawing and information visual- by the students. We also describe NetViz Nirvana, layout
ization. Our primary objectives were (1) to evaluate NodeXL
as an SNA tool for a broad base of users and (2) to explore principles that can increase the readability and interpretative
methods for teaching SNA. Our complementary dual case-study power of social network visualizations [3]. As a set of criteria
format demonstrates the usability of NodeXL for a diverse set aspired to by most of the students, we feel it is useful for
of users, and significantly, the power of a tightly integrated the future design of SNA tools in general and NodeXL in
metrics/visualization tool to spark insight and facilitate sense- particular. Finally, we offer lessons learned for educators,
making for students of SNA.
researchers, and developers of SNA tools such as NodeXL.
We report on the process of learning SNA techniques to study
I. I NTRODUCTION
online communities in [4].
The booming popularity of social networking is reflected
in recent statistics from the Pew Internet & American Life A. NodeXL Overview
Project [1]: the number of adult internet users who have NodeXL (Network Overview for Discovery and Explo-
a profile on an online social network site has more than ration in Excel) is an open-source SNA plug-in for Microsoft
quadrupled in the past four years, rising from 8% in early Excel 2007 (http://www.codeplex.com/NodeXL). NodeXL is
2005, 16% in 2006, to 35% in December 2008. Over half of intended to be easy to adopt for existing users of Excel,
those adults maintain more than two online profiles, generally taking advantage of common spreadsheet capabilities such
on different sites. They blog: 54% of college students read as sorting, filtering, and creating formulas. NodeXL extends
blogs and 33% write them; overall, 36% of adults read blogs the spreadsheet into a network analysis and visualization tool
and 13% write them, and almost 20% remix content into their by incorporating a library of basic network metrics (e.g.,
own inventions to be shared online [2]. The widespread use of degree, centrality measures, elementary clustering) and graph
social networking applications has precipitated a greater need visualization features. Data can be entered or imported into
by more diverse users to understand how their online commu- the NodeXL template and quickly displayed as a graph. It is
nities evolve and thrive. SNA tools are not just for scientists uniquely positioned to support the growing number of com-
anymore. Moderators, administrators and other community munity analysts who have neither the time nor desire to step
experts also have a stake in learning more about the structural through static visualizations or to learn complex programming
dynamics of their interactions. The emergent challenge for interfaces. Because it is an evolving open source project, users
designers and educators is to build easy to learn interfaces with programming skills can also access NodeXL compo-
that enable these users to discover community patterns and nents [5].

Fig. 1. The NodeXL Workspace. The dual pane view of network metrics (left pane) & network graph (right pane) provide an integrated snapshot of
statistics & visualization, along with built-in functions & controls that support exploration & discovery. Individual worksheets separate network analysis tasks
into separate categories, closely aligned with topology & attribute-based tasks outlined in [6], such as “Edges,” “Vertices,” “Clusters.” The social network
shown reflects voting patterns of U.S. senators, the analysis of which is detailed in [7] and [8].

Arguably, the most significant design model NodeXL im- mation. SNA tools and techniques, then, can be situated at the
plements is an integration of statistical measures and visual- intersection of social computing and visual analytics. While
izations (see Fig. 1). As noted in [7], “SNA is a deductive most SNA tasks can be classified according to a taxonomy
task, and a user’s exploratory process can be distracted by or general workflow ([13], [6]), the actual exploratory process
having to navigate between separate statistical and visual- is dynamic, non-linear, and dependent on contextual factors
ization packages” (p.266). In terms of interaction factors such as network size, complexity and aspects of the network
for visual analytics tools, this design feature is called the being investigated. Typical human-computer interaction tests
“connect” interaction [8]. This connectivity is manifest in the conducted in controlled lab-environments cannot adequately
dynamic linkage between the data spreadsheet view and the reflect the complexity of real-world SNA tasks.
graphic layout view. Activating one representation will cause a Many usability studies of visualization tools have been
simultaneous change in the other [9]. For example, a node and conducted in laboratory settings under controlled conditions
its adjacent edges are highlighted in the graph whether a user ([14], [15]). Conversely, ethnographic studies of technology
clicks on the visual representation of the node in the layout have allowed researchers to gain deep understanding of how
pane or the spreadsheet row for the node in the “Vertices” users interact with technological tools ([16], [17]) in context-
tab. The simultaneous views of network statistics and network dependant conditions [18]. While long term qualitative studies
graphs combine language and visual attention modalities such can be difficult to sustain and may present a challenge to
that each provides “cognitive offloading” scaffolds ([9], [10]), charting consistent usage patterns, they are more aligned
and can trigger associative learning [11]. with the ways users interact with complex analysis tools and
B. Related Work: SNA Tools’ Evaluation & Education engage in complex research problems. The MILCs approach
SNA practitioners explore complex sets of relationships is part of a growing movement in the development of visual
within social systems to discover codifiable patterns of inter- analytic systems that argues against the use of traditional
action or reveal structural signatures of social roles ([5], [12]). prescriptive, task-based evaluations ([14], [15], [7], [8], [19]).
Complex social relationships and community use of social The original MILCs aimed to: (1) help developers design and
media can become visible with SNA. Similarly, in the field refine information visualizations systems more effectively and
of visual analytics, complex, dynamic data sets are explored (2) spark expert users to unexpected insights and discoveries.
for unexpected insights and synthesized into actionable infor- Previous evaluations of visual analytic systems such as

SocialAction ([7], [13], [8]) and others (detailed in [20], ways students could study social interactions. This perspective
[19]) invited domain experts to evaluate the system. The situated SNA within an overall learning environment in much
same model can be applied to the evaluation of complex the same way MILCs have situated visual analytics tools
visualization systems used by teachers and students. Just as within the work environments of their domain experts.
exploratory efforts in scientific inquiry can be mapped to a Before instruction, IS users’ level of Excel and SNA lit-
cycle of sense-making tasks with feedback at each level [21], eracy was assessed by a self-reporting questionnaire. Most
exploratory learning is a cycle of sense-making tasks with students reported a basic competency level in Excel and almost
multiple feedback loops [11]. An early example of the MILCs none of them had any prior knowledge of SNA. Their first
paradigm in educational contexts is detailed in [22]. Of note, encounter with NodeXL was a tutorial session conducted
its non-structured observation format empowered students to in a computer lab, following the outline and examples of
explore, and in many cases reach new insights (“discovery Network Analysis with NodeXL: Learning by Doing, a draft
learning,” [22]). A main advantage of extending the MILCs introduction to SNA text written by 3 of the authors. The
approach to learning contexts is that it offers students the tutorial tasks supported 4 of the 7 workflow steps outlined
same evaluation and design affordances it holds for research in [13], and reflected the general topology- and attribute-based
with domain experts: the MILCs process reflects the real tasks presented in [6]’s task taxonomy for graph analysis.
way students explore an interface as they are learning SNA During the study, participants engaged in weekly class discus-
concepts. sions and completed 3 assignments between classes. The first
In related work on SNA education, [23] summarized the assignment was based on a pre-defined dataset while follow-
results of a course on optimizing the online communication on assignments allowed participants to independently explore
behavior of student teams collaborating across geographic SNA-related research questions on their own datasets. The
distances while studying SNA. Insofar as one-third of the students were observed as a group during class sessions, but
course presented social network analysis concepts, its format each also had a 1-hour one-on-one session with a researcher.
resembled the seminar we studied. In [23]’s case, the students The individual sessions followed a contextual inquiry protocol,
collaborated on course work while applying SNA principles in which participants ’talk-aloud’ as they work, while the
to evaluate their own communication patterns. In contrast, our observer asks clarifying questions and makes note of important
intent was to evaluate how an interface like NodeXL can be occurrences [24]. Self-reporting diaries are an integral part of
used as both an SNA tool and a means to teach SNA. the MILCs approach [15], and also allowed participants to
capture and reflect upon their actions, emotions, and moments
II. M ETHODOLOGY – T WO - PRONGED A PPROACH of insight ([25], [26]).
As noted in §I-B, we used a qualitative approach, based pri-
B. Methodology: CS users
marily on MILCs [15]. MILCs employ multiple user analysis
methods within the same case study to attain a rich evaluation We used similar techniques as detailed above (§II-A) for 6
of the researched tools. In order to conduct an evaluation CS students, compressed into a single 90 minute session for
of NodeXL sensitive to diverse categories of users (IS & each. Two self-reporting questionnaires were given to each
CS students), we designed a two-pronged appraisal approach. participant before any instruction took place: one a subset of
For each group of users we used a core set of instruments the Excel and SNA literacy survey taken by our IS users,
and techniques: self-reporting pre-survey, tutorial session with and the other part of the standard NodeXL network analysis
NodeXL; observation/interview, and post-survey. However, the survey optionally taken by new users of the tool. The latter
length and depth-of-focus of our evaluations were tailored to deals more specifically with the user’s experiences with other
the background characteristics of each group. Because the IS network analysis tools and their academic background. As
users were using NodeXL as part of a semester-long course, they were more adept with both SNA tools and techniques,
we could follow the MILCs model with them, making in-situ we favored one-on-one tutorials with the CS users in lieu of
observations over time and allowing them to reflect on their class instruction. This format also allowed them more freedom
use of the tool as they grew more proficient. We used a more to ask questions and critique NodeXL as new features were
compressed but focused evaluation for the CS users, as they introduced. The tutorial followed the same outline used by the
required significantly less time to learn NodeXL. Although the IS users. In the second half, the CS users were asked to select
detailed methodologies employed for each group differed, we a dataset of interest to them out of the ones we had available or
believe our two-pronged approach enabled a rich, thorough their own collection and to identify and highlight three things
evaluation of NodeXL as a tool for a broad base of users. within it:
• a unique social role within the network,
A. Methodology: IS users • an interesting sub-group within the larger network, and
Over a 5 week period, 15 IS graduate students participating • anything else of particular interest.
in a “Communities of Practice” seminar were introduced to During their analysis, which lasted 30-45 minutes, we
SNA and used NodeXL to explore SNA fundamentals. While observed each participant individually in the same manner
SNA was presented as a powerful set of techniques for analyz- as the IS users. Their comments, along with each step they
ing online communities, it was considered only one of many took in their analysis and any problems they encountered were

However, a tool that enables unrestricted manipulation of
these attributes can serve as a double-edged sword. While
it allows users to create complex visualizations of multi-
dimensional data, over-use of multiple display options can
quickly reduce the graph’s readability and obscure potentially
significant perspectives. To ensure that students maintained a
high awareness of the importance of producing readable graphs
that accurately reflect the community relationships they hoped
to convey, the students read and discussed the readability
metrics (RMs) outlined in [3]. They were enjoined to aspire
to the four principles of NetViz Nirvana [3]:
• Every node is visible
• Every node’s degree is countable
• Every edge can be followed from source to destination
• Clusters and outliers are identifiable
In-class and online discussions about NetViz Nirvana, cou-
pled with NodeXL’s tight integration of statistics views and
graph layouts empowered the students to produce relatively
Fig. 2. Student Product: The friendship relationships of an online com- sophisticated SNA graphs in a short period of time. Most
munity. The visualization uses multiple SNA metrics to reveal structure: students took full advantage of the ease with which NodeXL
node color reflects clustering coefficients (identifying possible subgroups and
their connectivity to the whole); edge colors reflect geographic connections. can map individual and collective metrics to the shape, size,
Node radius reflects betweenness centrality (i.e., bridges/boundary spanners and color attributes of visual graph objects. A sample of
for the community overall), and shape reflects either closeness centrality or student products, along with captions detailing their layout
eigenvector centrality (no nodes met both metrics at the same time). Those
members with high closeness centrality are triangles; those with low closeness choices, are shown in Figures 2–5. In each case, the way
centrality are circles; those connected to the most popular members are NodeXL supports thoughtful use of shape and color, as well
squares. as selective labeling of nodes of interest, helps direct attention
to important relationships.
The student who created Fig. 2 used multiple SNA metrics
noted by the observer so as to form an approximation to the to reveal her community’s structure (see caption for details).
self-reporting diaries used for the IS users. Personal diaries The layout she chose also follows many NetViz Nirvana
and class assignments were omitted due to the experience design principles, as larger, more connected nodes are pulled
and technical background of the CS users, as well as the into the layout’s center. The student who created Fig. 3
more time limited nature of the study. However, this one-on- focused on identifying the relationship between 3 community
one perspective afforded us an increase in detailed responses forums. Like the Fig. 2 author, this student also mapped
from each individual. At the conclusion of the second half metrics (degree & tie strength) to visual properties (diameter
participants completed 2 additional questionnaires: one being & edge thickness), using her graph results to identify boundary
the remainder of the standard NodeXL survey mentioned spanners in the community and highlight a close connection
above that deals with the user’s experience with NodeXL and between two sub-groups. Again, the use of shapes and color
the other select questions from the self-reporting diaries used were used to draw attention to important individuals, supported
for the IS users. by the ability to label only select nodes. Another student
III. N ODE XL & SNA S ENSE M AKING applied a simple NodeXL “skip” filter to model a complex
community management dilemma: how to find the best can-
As noted in [3], Euclid established that a point is that didate to replace a departing administrator (Fig. 4 & 5). He
which has position but no magnitude. He also defined a line hypothesized that his admin and experienced community mem-
as that which has position and length, but no breadth [27]. bers would have higher eigenvector and betweenness centrality
While these axioms are applicable in plane geometry and measures, and reflected these metrics in visual properties.
algorithms historically used to display graphs [3], nodes and Using NodeXL’s “skip” filter to remove the existing admin, he
edges in social networks must convey the same attributes as was quickly able to model and “see” which member possessed
the people and relations they represent. Practitioners who wish the best network values to take over (see captions for details).
to explore community member characteristics such as number
of contributions, gender, or degree (i.e., number of friends) IV. A SPECTS OF N ODE XL U SABILITY
may want to represent them visually via multiple node and Of our 6 CS users, 5 had a course background in network
edge attributes such as shape, size and color. The ability to visualization, and 4 also had general information visualization
apply a variety of visual features to SNA data elements based course experience. Only one participant lacked both. Three
on various individual and community metrics was a NodeXL of the participants had studied SNA as part of a course, and
feature that all students used and enjoyed. one has presented network analysis findings in both academic

updating, instead preferring to make all the data modifications
beforehand and visualizing the data only as a final step.
Conversely, this real-time updating was the most appealing
aspect to 3 of our users, who also requested that “autofill
columns” immediately update the graph pane with the results
instead of only placing them in the workbook. This feature,
as well as the optional disabling of real-time updating, has
been added to NodeXL. Though it would further increase the
response time, 2 users also asked that the autofilled columns
to optionally stay in sync with their source data.
Another key complaint is the desire for increased usability.
Our participants expected many operations to be available
directly from the context menu of the graph pane, including
fixing vertex placement, deletion of vertices and edges, and
skipping filtered out vertices and edges after using dynamic
filters. Further, users found the NodeXL ribbon tab to be
confusing and too detailed, instead requesting simpler, larger
Fig. 3. Student Product: This graphic helped the student identify key icons that give precedence to frequently used features. The
boundary spanners (e.g., ateixeira), recognize the close connection between 2 latest version of NodeXL has implemented part of an interface
different forums, and the lack of significant activity by the official community
leaders. Blue triangles are community leaders; green diamonds are hosts. She overhaul designed to address these issues, with basic features
could also easily label nodes selectively. However, she did have to manually exposed in the ribbon tab and grouped in a more orderly
create a separate legend to explain these roles and relationships. fashion while advanced ones are moved to pop-up dialog
boxes. CS users also requested shortcuts for many existing
functions, some of which we have already implemented.
and business settings. Thus, with CS users, an emphasis was For some datasets, users hit the limits of edge aggregation
made on specific usability aspects of the tool. However, these within NodeXL. Two of our users requested more expressive
aspects were also correlated against IS users’ feedback. We aggregation techniques than only a count of the number of
categorized the CS user surveys and comments according aggregated items, instead preferring the option of selecting a
to the tutorial outline described in §II-B, with additional measure such as sum, median, or mean for the weight column.
areas added as needed for comments that defied classification. Another user wanted a way to easily merge vertices while
Overall, we found 16 unique acclamations of NodeXL across aggregating the data columns.
10 aspects of the tool, 5 of which were were outside the Finally, when our participants wanted to output the results of
scope of the tutorial document. Moreover, we found 96 unique their analysis they found the existing image export features of
criticisms and feature requests across 37 areas, which were NodeXL lacking. Specifically, many requested vector graphic
substantially more specific than the acclamations. 17 of the formats such as SVG, EPS, and WMF to output publishable
37 areas were mentioned during the IS users’ tool exploration, images. We added an XPS export to satisfy this need. To im-
not instruction. Of the 16 acclamations, 3 were made by 2 or prove text legibility and reduce file size they suggested a com-
more of the participants. pressed, lossless format like PNG replace NodeXL’s default
The interaction between the workbook and the graph pane in image export options (JPEG/lossy and BMP/uncompressed).
NodeXL was quite appealing to the CS users, as they wanted to All participants rated their satisfaction with NodeXL higher
maintain a separation of data and visualization while enabling than current network analysis software except for one, who
connections through brushing and linking. One user especially rated NodeXL equivalent to current software. This participant
liked the ability to use cell referencing within the worksheets had the weakest background in Excel, SNA techniques, and
to define tunable constants for their vertex attribute formulas, alternative SNA tools among those studied.
something he had not been able to do in other tools. Of the
criticisms and feature requests, 11 were brought up by at V. L ESSONS L EARNED FROM VARIOUS P ERSPECTIVES
least 2 participants and 2 were mentioned by at least 3 of The process of evaluating visual analytics tools involves
them. The foremost concern our CS users had was the lack the intersection of learning a new, complex system, learning
of responsive controls at all times during their analysis. While the language of complex SNA concepts, and applying these
running the layout algorithms, 3 users requested a progress components to abstract goals. How do you learn to explore
indicator with stop or pause buttons, even though the graph and interpret patterns of interaction in online communities to
pane is updated after every iteration. Moreover, the brushing reveal latent social structures or signature roles? [12] Our study
and linking between the workbook and visualization slowed provides a first look at the impact a tool like NodeXL may have
down 2 users who worked with large datasets, as selecting on the teaching and learning of fundamental SNA concepts,
many vertices in either caused a substantial update in the other. and the benefits of using MILCs as an evaluation method
One participant even requested the ability to disable real-time for the learning process. Our two-pronged methodology for

into one integrated visual analytics tool was effective in
empowering new users of SNA tools to create quite sophisti-
cated, meaningful social network diagrams. Most students took
advantage of the close coupling between spreadsheet metrics
and the visualization to manually manipulate their layouts,
moving fluidly between 1) finding a node in the spreadsheet
to maneuver it on the graph pane, and 2) selecting a node
in the graph to review statistics for it from the spreadsheet.
This interaction was cited by the CS users as a key feature
of NodeXL. The Communities of Practice instructor found it
useful during teaching to highlight SNA concepts as they were
presented on the classroom projector. However, the “connect”
interaction paradigm designed into NodeXL did not extend to
the automatic creation of labels or legends for visualizations
created by users – a feature that is standard in chart features
found within Excel. Most students had to create their own
legends for nodes (see Fig. 3). This failing has been addressed
in an upcoming release.
Fig. 4. Student Product: The student confirms a hypothesis that the
Filter: NodeXL’s suite of “filter” interactions seems to be
community administrator and more experienced members will have high missing a classic “details-on-demand” feature. Several students
eigenvector centrality (connectedness, represented by node radius) and be- commented that in order to make side-by-side comparisons
tweenness centrality (bridging, represented by node color). He takes this idea
one step further in Fig. 5.
of columns of interest in the worksheets, they had to scroll
back and forth constantly. At times, their efforts to cut-and-
paste user-created columns near built-in columns to simplify
comparisons caused unexpected results (either human error in
the “cut-n-paste,” or a lack of awareness about the inability
to mix certain types of columns in NodeXL). Based on
informal feedback with NodeXL developers, such straight-
forward comparisons are possible, but require multiple steps.
A macro could be created as a shortcut, following the familiar
“cut-n-paste” paradigm. Or, in some cases, the “freeze panes”
feature could be used more effectively.
Encode: The “encode” interaction, or the ability to apply
shapes, size, labels, orientations by encoding data elements
with attributes of interest, is the most popular way in which
students chose to represent their communities. NodeXL offers
an impressive array of visual attributes to represent SNA
metrics. Variations in node/edge color, size and shape were
used extensively by students to improve the interpretive and
expressive power of their visualizations. Most students used
relatively simple metrics like degree for nodes and tie-strength
Fig. 5. Student Product: Modeling which node would be the next-best for edges. Five of the 12 IS users felt their use of extensive
candidate to serve as community administrator. To simulate this management metrics was limited due to the structure of the data they
dilemma, statistical data about the current admin was ’skipped’ via NodeXL
filtering functions, & the next best “node” pops up in his visualization. collected – many of the students chose to collect data about
community member’s connections to community artifacts,
such as contributions to various forums (see Fig. 3), and in
studying different types of users has also enabled us to offer one case, time-stamps. In such cases, metrics are not easily
a robust evaluation of NodeXL as a visual analytics tool. The interpreted or may not apply.
guidelines for interaction design proposed by Ji Soon Yi (cited Explore: The students enjoyed using attribute rankings to try
in [8]), such as “connect,” “encode,” “filter” and “explore” to make their graphs more readable. However, in their effort
serve as our evaluation criteria, as described in the following to reach NetViz Nirvana, they still had to invest in a great deal
lessons learned. of manual manipulation. As noted in [3], automated layout
algorithms are not sufficient to effectively represent the range
A. Lessons Learned for Designers of attributes and often localized structures that most SNA
Connect: The “dual-front approach,” or “connect” inter- experts and students would like to highlight. Based on the
action design [8] that combines statistics and visualizations exploration process followed by many of the users, we believe

the addition of specific RM interactions into the design of SNA semi-automatic “Calculate Metrics” function, a powerful li-
tools would be of great value. Specifically, node RMs such brary of basic techniques, often hid some of the instructional
as node occlusion and node size, shape, and color variance power of statistics from the students. Interaction functions to
constraints should be the highest priority (a potentially positive allow metric tweaking and graph response, along with interac-
side effect of incorporating these node RMs is they may tive RM layout functions could enable “teachable moments”
reduce the risk of edge tunnels). An interactive edge RM that make SNA metrics explicit, guide the user through rec-
for reducing edge crossings would also be of immediate ommended steps required to attain NetViz Nirvana, and allow
value, as many students used edge thickness to reflect tie- further opportunities for experimentation and exploration to
strength. Both IS and CS users alike had difficulty using labels reach novel insights.
effectively, underscoring the potential benefits of node RM Tutorial tasks did not take advantage of the “reconfigure”
research efforts. interaction design [8], i.e., offering different perspectives on
The ability to help users track their exploration process, or the same data. NodeXL is embedded within Excel, which is
navigate through a history of their actions is also a component equipped with multiple existing charts that can be exploited
of the “Explore” interaction. Currently, NodeXL does not to display data through different lenses (e.g., scatterplot,
contain any supports for “undo” or traceable histories of histograms). Emphasis was placed on creating and analyzing
exploration or annotation paths. Students were quite frustrated network views or sorting the tabular spreadsheet to find
with the interruptions they faced when the tool crashed, and interesting patterns. Time constraints within the course, along
often, “had to start over.” SocialAction is one SNA tool that with student lack of in-depth familiarity with Excel’s features
can serve as an example in adding these features to the design reduced opportunities to experiment with and learn from a
queue [13]. An “Undo/Redo” function would appear to be greater variety of perspectives.
especially useful in beta-testing contexts, when users who The ability to keep track of actions is a scaffolding sup-
experience multiple system crashes are supported by some port that enhances learning, confidence and enables extensive
means for reverting back to a recent “safe” mode. (Note: a freedom to explore. As noted in §V-A, an “undo” feature for
few users successfully used Excel’s crash recovery feature to adjustments made to either the spreadsheet view or the layout
resume analysis.) of nodes should be added. A work-around several students de-
veloped for lack of this function was to keep track of multiple
B. Lessons Learned for Educators/Tutorial Designers saved versions of their work, or maintain various simultaneous
NodeXL was found to be a relatively easy to learn tool. Excel spreadsheets that reflected their workflow status. These
Excel experience and familiarity with graphs and networks additional files required explicit tracking, which may have
appear to be baseline skills for effective use of NodeXL. Lack increased the cognitive load the students experienced.
of in-depth knowledge of Excel proved to be the primary
barrier to the learning process for IS users. Three of the C. Lessons Learned for Researchers
CS users and 5 of the IS users felt their Excel experience MILCs are indeed difficult to execute and analyze [28], due
was useful, while 8 of the IS users felt their inexperience to amounts of data collected, and the amount of time invested
with Excel, especially a lack of expertise in creating Excel by both participants and researchers. However, they provide
formulas, prevented their rapid adoption of the full features of more accurate representations of the processes experts and
NodeXL. learners follow while exploring complex data sets, and can
Pacing issues, both in the tutorial task flow, lab sessions, and result in more meaningful evaluation measures and recommen-
assignments, were of paramount concern to the IS users, who dations for improvement in design and teaching approaches.
were asked to learn the language of SNA, common statistical Exploratory learning can be described, fundamentally, as a
and graph-based approaches, and a new tool in a relatively cycle of exploration, insight (learned concept), and deeper
short amount of time. Users had less than 2 weeks to learn the exploration. Consequently, it is difficult to shoe-horn into
basics of NodeXL and relevant SNA metrics, then to interpret specific tasks – much like the exploratory analysis processes
their first set of graphs, starting with a pre-defined data set followed by experts. MILCs are thus an effective, rich infor-
that did not reflect their communities of interest. Nine of 12 mation source for visual analytics usage and learning. Overall,
IS users noted that the tutorial and first assignment made too this study not only adds to the growing body of research
steep and quick a transition from small networks of relatively establishing MILCs as an effective evaluation method for
low graph density to large networks of high graph density. visual analytics tools, but also extends it as a valid approach
Similarly, our CS users emphasized the importance of a series for evaluating systems designed for exploratory learning,
of small tasks with predefined goals, especially when the users specifically those designed for a wide range of learners.
do not have a vested interest in the dataset [28]. Previous MILCs and a field study on exploratory learning
Promoting user awareness of layout considerations such as ([19], [26]) captured moments of insight via “Eureka” reports.
NetViz Nirvana resulted in a high level of reflective thought In these early studies, users not only recorded the analysis
and effort put forth by IS users in their analyses. Thus, steps they followed, along with times-on-task for major activ-
incorporating NetViz Nirvana functions such as real-time RM ities, but also reflected upon key insights they experienced. In
interactions may support educational goals for SNA. NodeXL’s future research applying MILCs to teaching/learning contexts,

explicitly capturing similar “Eureka” reports may prove of [5] M. Smith, B. Shneiderman, N. Milic-Frayling, E. M. Rodrigues,
value. V. Barash, C. Dunne, T. Capone, A. Perer, and E. Gleave, “Analyzing
social (media) network data with NodeXL,” in C&T ’09: International
VI. F UTURE W ORK AND C ONCLUSIONS Conference on Communities and Technologies 2009, 2009.
[6] B. Lee, C. Plaisant, C. S. Parr, J.-D. Fekete, and N. Henry, “Task
In their post-session survey, CS users prioritized pending taxonomy for graph visualization,” in BELIV ’06: Proc. 2006 AVI
NodeXL development tasks in the following order: (1) better workshop on BEyond time and errors: novel evaLuation methods for
Information Visualization. ACM, 2006.
automated layout algorithms, (2) clustering algorithms or [7] A. Perer and B. Shneiderman, “Integrating statistics and visualization:
additional means to group nodes, and (3) additional metric Case studies of gaining clarity during exploratory data analysis,” in
algorithms to measure networks. Other options were more CHI ’08: Proc. 26th annual SIGCHI conference on Human Factors in
Computing Systems. ACM, 2008.
evenly tied, though, interestingly, all of our CS users ranked [8] ——, “Integrating statistics and visualization for exploratory power:
3D layouts last on their list. IS users were also interested From long-term case studies to design guidelines,” IEEE Computer
in improved layout algorithms and struggled with a need for Graphics & Applications, no. May/June, pp. 24–36, 2009.
[9] K. Sedig and H.-N. Liang, “Interactivity of visual mathematical repre-
better support of grouping nodes. sentations: Factors affecting learning and cognitive processes.” Journal
The top design suggestion made in addition to specific of Interactive Learning Research, vol. 17, no. 2, pp. 179–212, 2006.
survey responses was the ability to use NodeXL outside of [10] C. Ware, Visual thinking for design, ser. Morgan Kaufmann Series in
Interactive Technologies. Morgan Kaufmann, 2008.
Excel 2007. Both CS and IS students requested a means to [11] S. d. Freitas and T. Neumann, “The use of ‘exploratory learning’ for
use NodeXL outside of Excel 2007. Several users desired supporting immersive learning in virtual environments,” Computers &
a port to OSX, while another requested a completely open Education, vol. 52, no. 2, pp. 343–352, 2009.
[12] H. Welser, E. Gleave, D. Fisher, and M. Smith, “Visualizing the
source implementation. NodeXL is released under the Mi- signatures of social roles in online discussion groups.” Journal of Social
crosoft Public License, and a Silverlight implementation may Structure, vol. 8, 2007.
allow for broader adoption. Much of the other users feedback [13] A. Perer and B. Shneiderman, “Systematic yet flexible discovery: Guid-
ing domain experts through exploratory data analysis,” in IUI ’08: Proc.
not mentioned here has already been included in the current 13th international conference on Intelligent User Interfaces. ACM,
release. 2008.
Our findings demonstrate the power of the complementary [14] B. Kitchenham, L. Pickard, and S. L. Pfleeger, “Case studies for method
and tool evaluation,” IEEE Software, vol. 12, no. 4, pp. 52–62, 1995.
methodologies we used to evaluate NodeXL as a tool for SNA [15] B. Shneiderman and C. Plaisant, “Strategies for evaluating information
experts and teaching: visualization tools: multi-dimensional in-depth long-term case studies,”
• our user pool represented both diversity & depth from a in BELIV ’06: Proc. 2006 AVI workshop on BEyond time and errors:
novel evaLuation methods for Information Visualization. ACM, 2006.
research methods perspective; [16] M. Tory and S. Staub-French, “Qualitative analysis of visualization:
• our IS users’ feedback showcased NodeXL’s power as a a building design field study.” in BELIV ’08: Proc. 2008 conference
teaching/learning tool for SNA; on BEyond time and errors: novel evaLuation methods for Information
Visualization, ACM, Ed. ACM, 2008.
• our CS users’ feedback enabled us to compare NodeXL [17] J. Betteridge, “Answering back: The telephone, modernity and everyday
to existing tools & enabled us to rapidly implement life.” Media Culture Society, vol. 19, no. 4, 1997.
requested features & fixes during the study & beyond. [18] B. Kaplan and D. Duchon, “Combining qualitative and quantitative
methods in information systems research: A case study.” MIS Quarterly,
Our findings suggest that NodeXL enabled sharp, but non- vol. 12, no. 4, pp. 571–586, 1988.
technical students to interpret and create meaningful represen- [19] P. Saraiya, C. North, V. Lam, and K. Duca, “An insight-based longitudi-
nal study of visual analytics,” IEEE Trans. Visualization and Computer
tations of complex social structures in a fairly short timeframe. Graphics, vol. 12, no. 6, pp. 1511–1522, 2006.
Their success was based on the tool’s close coupling of [20] P. Saraiya, C. North, and K. Ducak, “An evaluation of microarray
spreadsheet and graph visualization, as well as the range of visualization tools for biological insight.” in INFOVIS ’04: Proc. IEEE
Symposium for Information Visualization. IEEE, 2004, pp. 1–8.
ways it allows metrics and attributes to be mapped onto graphs. [21] J. Thomas and K. Cook, Illuminating the path: The research and
As a result, nearly all of our users found NodeXL to be more development agenda for visual analytics., 2006.
usable and learnable than existing SNA tools. [22] G. Marchionini and G. Crane, “Evaluating hypermedia and learning:
Methods and results from the Perseus Project,” ACM Trans. Information
ACKNOWLEDGMENTS Systems, vol. 12, no. 1, pp. 5–34, 1994.
[23] P. A. Gloor, M. Paasivaara, D. Schoder, and P. Willems, “Finding
We thank the University of Maryland participants who collaborative innovation networks through correlating performance with
offered improvements for NodeXL. Natasa Milic-Frayling social network structure,” International Journal of Production Research,
vol. 46, no. 5, pp. 1357–1371, 2008.
provided helpful suggestions and support. Microsoft External [24] K. Holtzblatt and S. Jones, Contextual inquiry: A participatory technique
Research provided partial support for this project. for system design. Lawrence Earlbaum, 1993.
[25] S. Carter and J. Mankoff, “When participants do the capturing: the role
R EFERENCES of media in diary studies.” in CHI ’05: Proc. SIGCHI conference on
[1] A. Lenhart, “Adults and social network websites,” Pew Internet and Human Factors in Computing Systems, ACM, Ed., 2005.
American Life Project, Tech. Rep., Jan 14, 2009 2009. [26] J. Rieman, “A field study of exploratory learning strategies,” ACM Trans.
[2] L. Rainie, “The new news media-scape,” Feb 17 2009. Computer-Human Interaction, vol. 3, no. 3, pp. 189–218, 1996.
[3] C. Dunne and B. Shneiderman, “Improving graph drawing readability by [27] C. L. Dodgson, Euclid. Books I. II, 1882.
incorporating readability metrics: A software tool for network analysts,” [28] R. A. Valiati, Eliane, C. M. D. S. Freitas, and S. Pimenta, Marcelo,
University of Maryland, HCIL Tech Report HCIL-2009-13, May 2009. “Using multi-dimensional in-depth long-term case studies for informa-
[4] D. L. Hansen, D. Rotman, E. M. Bonsignore, N. Milic-Frayling, E. M. tion visualization evaluation,” in BELIV ’08: Proc. 2008 conference on
Rodrigues, M. Smith, and B. Shneiderman, “Do you know the way to BEyond time and errors: novel evaLuation methods for Information
SNA?: A process model for analyzing and visualizing social media data.” Visualization. ACM, 2008.
Submitted, 2009.

2009-Social computing-First steps to netviz nirvana

More Related Content

What's hot (20)

Similar to 2009-Social computing-First steps to netviz nirvana (20)

More from Marc Smith (20)

Recently uploaded (20)

2009-Social computing-First steps to netviz nirvana