Harry Potter and the Meat-Filled Freezer: A Case Study of Spontaneous Usage of Visualization Tools

Harry Potter and the Meat-Filled Freezer:
A Case Study of Spontaneous Usage of Visualization Tools
Fernanda B. Viégas, Martin Wattenberg, Matt McKeon, Frank van Ham, Jesse Kriss
IBM T. J. Watson Research Center
Abstract
This paper is a report on early user activity in Many
Eyes, a public web site where users may upload data,
create visualizations, and carry on discussions. Since
the site launched, users have uploaded data and
created graphics on everything from DNA microarray
data to co-occurrences of names in the New
Testament to personal gift-giving networks. Our
results show that in addition to traditional data
analysis, Many Eyes is used for goals ranging from
journalism and advocacy to personal expression and
social interaction. We propose several implications of
this usage for visualization designers and contend that
these findings suggest a role for visualization as an
expressive medium.
1 INTRODUCTION
The biggest words on the visualization are
chicken, pork, pulled, steak, and thighs. Beef,
salmon, and tenderloin also stand out in large
type. On closer inspection, you realize that
pastrami, prosciutto, and turkey are buried there
too. Looking at the title, you see that this is a
visualization of the contents of John’s freezer—
only the meat, of course. You don’t know John
and you may never meet him but, at some
unexpected level, you can relate to the contents of
his freezer.
Figure 1: Tag cloud created on Many Eyes
This tag cloud was created by a users of Many
Eyes, a website where people may upload data,
create visualizations, and carry on discussions.
The site provides tools for collecting, visualizing,
and discussing data. (The system design is
described fully in [18].) In this paper we examine
early user activity on the Many Eyes site with an
eye toward answering two main questions. The
first is focused: Do usage patterns corroborate or
contradict the hypotheses advanced in [15] [20] [6]
that visualizations can catalyze social activity?
The second question is open-ended: When
powerful visualization technology is made
broadly available, how do people spontaneously
apply it? Do they engage solely in the visual
analytic tasks that are conventionally associated
with complex information displays, or do they use
visualizations for other purposes?
The first question, regarding social activity, is
prompted by results that indicate visualization
may have a catalytic effect on communication
between users. The evidence for this hypothesis,
however, is still incomplete. In the case of
PostHistory [15] and the NameVoyager [20],
social interaction was observed around
visualizations, but in each case it was an
unexpected side effect of a system designed for
other purposes. In the case of Sense.us [6], lab
studies and limited deployments provided further
evidence, but data on extended public usage was
not available.
The second question, on how people
spontaneously apply visualization when given free
rein, arises because—for many viewers of Many
Eyes—the site represents the first time they have
had the ability to make their own visualizations.
For the most part, information visualization
research projects are rarely available to a large
population of users. Typically researchers build
applications with specific data sets in mind and it
is not uncommon for the creators of a
visualization system to be its main users. Even
when user studies are carried out, the setting is
usually controlled and the tasks are often predetermined.
Artificial tasks and environments
make for repeatable experimental results, but
possibly at the cost of general validity. It is hard to
observe visualization usage “in the wild” and it
may be even harder to find a large body of users
to study.
To address these issues, proposals have been
made for a “case study” approach to visualization
evaluation [6]. Learning how users employ
applications seems essential for progress in the
field of information visualization. Studying the
ways in which everyday usage differs from the
intended goals of the tools we build might open
new areas of research for the infovis community.
For example, visual systems are typically built on
the assumption that data exploration and analysis
are the main tasks people want to accomplish
when using a visualization application. How true
is this belief? It is hard to say, given that we do
not have a large body of accounts on how users
utilize visualization tools.
Because Many Eyes is a public online system,
we have the rare opportunity to see how a
visualization application fares with thousands of
users. Since our target audience is regular Web
users (anyone who knows how to browse the Web
pages, and has some sort of interest in data), Many
Eyes is able to attract a wealth of visitors ranging
from novices, who may never have played with
interactive visualizations before, to data experts
such as scientists and journalists.
To examine the activity of this set of users we
employed a range of methodologies, from
standard empirical methods—content coding and
statistical analysis—to less conventional
approaches such as using search engines to find
blog posts that referred to Many Eyes. We studied
site activity in all its different manifestations: data
sets uploaded, visualizations created, and
comments contributed.
The results suggest a strong social component
of Many Eyes usage, ranging from playful
behaviour (using visualization as games) to
socially conscious activity (visualization for
advocacy and solace) to turning visualizations into
a mechanism of self-identity (data mirrors). In
answer to our second question, we find that data
exploration and analysis are not the only goals
when users view and create visualizations. Indeed,
a data set that describes freezer contents is just as
characteristic as a table of numbers on global
warming.
The next section presents an introduction to
Many Eyes and an overview of related work. We
then introduce case studies that illustrate the ways
in which Many Eyes visualizations are used on
and off the site. The last sections describe content
coding results and an analysis of the limitations of
this study. We end with a discussion of the
implications of our findings.
2 BACKGROUND AND RELATED WORK
2.1 About the Many Eyes site
Many Eyes is described in detail in [18]; here
we only provide a summary of the aspects
important for this paper. The site launched in
January of 2007, is freely available to the public,
and provides “visualization as a service.” Users
can upload data to a shared public repository, and
then apply any of fourteen basic display
techniques—ranging from pie charts to
treemaps—to those data sets to create new
visualizations.
The site is designed for social interaction as
well as visualization. Each data set and
visualization has an associated discussion area
where users can place comments. When a user
places a comment on a visualization, a link is kept
to the state of the display at that moment; a user
reading that comment can click on the link to
restore the visualization to the state it was in when
a viewer saw it. The mechanism is similar to those
used in Spotfire DecisionSite Posters [12] and
DEVise [7].
A distinctive aspect of Many Eyes is that it has
been designed to exist as part of the Web
ecosystem. Users can place images and links to
the site on their blogs, with a special button
providing the necessary HTML code. Feeds in
RSS format are available to alert viewers to new
data, visualizations, or comments of interest. As a
result, a significant amount of activity related to
Many Eyes occurs on other web sites.
2.2 Related work
Several streams of investigation are relevant to
the current work. In the past decade much
attention has been paid to the area of social
visualization: the display of social data for social
purposes [15]. By this definition, some of the
practices that have spontaneously arisen on Many
Eyes may be considered examples of social
visualization. Nonetheless, Many Eyes as a whole
differs from conventional social visualization
systems in two respects. Social visualizations have
traditionally been aimed specifically at data from
a particular system (e.g. Usenet or chats) rather
than providing generic visualization components,
and they often are meant to serve as a conduit for
communication or impression formation rather
spurring discussion in a separate forum or context.
A second related area is a set of
“communication-minded” visualizations [17] that
have been reported and created in recent years.
Reports on systems such as the NameVoyager,
Vizster, and Sense.us have explored how
visualizations can be a spur to social, playful
analysis of data. Many Eyes was designed in part
to provide a large-scale testbed of these ideas.
In addition to serving as a testbed for the social
aspects of visualization, Many Eyes provides a
platform to examine broader questions about
visualization usage. Several authors have
addressed the question of what people want from
visualization. One approach, e.g. [1], has been to
ask potential users what they would want to know
about various data sets. Another approach is
described by [9] in which Plaisant notes that case
studies of tools in realistic settings form the least
common type of study—yet it is potentially one of
the most useful. The current paper presents such a
study with a wide base of users on the Internet.
Finally, several commercial web sites aimed at
data sharing and charting have recently appeared.
Swivel [13], Data360 [3], and Dataplace [4] all
provide web-based repositories where users can
upload data and make simple charts. All are aimed
at group exploration of data, though none provides
the kind of sophisticated visualizations seen on
Many Eyes. No studies have been published on
usage patterns of these sites.
3 QUALITATIVE ACCOUNT
Users create visualizations on Many Eyes for a
diverse set of reasons. Aside from simple testing
of the site, we have seen examples that range from
scientific research to artistic expression. In this
section we touch on the major themes that have
emerged, with in-depth descriptions of examples
of each. The qualitative descriptions are meant to
be miniature case studies that provide a sense of
the richness of site activity. In several cases, we
have included additional information gleaned
from blog entries that discuss particular
visualizations. In Section 4 we augment these
descriptions by quantitative analysis.
Figure 2: Global warming visualization and discussion
3.1 Visual Analytics—global warming and
blogs
Three days after the site launched, user Phil1
uploaded a data set entitled 420,000 Years of CO2
Levels and Temp and used a scatterplot to display
it (Fig. 2). The visualization has generated 20
comments as of this writing. The comments range
from political arguments to statistical analysis.
The interest in the data set is clearly not just
academic, but comes from the context of scientific
reports and news coverage about global warming.
As Phil, the uploader of the data, commented:
Interesting indeed, the collaborative
manipulation of data like this is really something!
I hadn't tried those arrangements, and now it
shows something else quite interesting. […] What
does this mean though that such high CO2 levels
are preceding significant changes in temperature?
Something, much worse is yet to come?
A different kind of analytic activity comes
from user “Elton”, a PhD student in Linguistics in
Germany. (These details are known because he
writes about them in a blog.) His research looks at
corporate blogs as a new genre with distinctive
linguistic characteristics. After stumbling upon
Many Eyes, a series of visualizations of his data
corpus have become additional evidence for his
thesis work. Elton has, thus far, uploaded 43 data
sets that track formality measurements on blog
entries. He has visualized the majority of these
data sets (sometimes using more than one
technique for each set of data). The activities of
Phil and Elton show that Many Eyes is used for
traditional data analysis.
3.1.1 Data Quality and Error Hunting
One of the main concerns in a site like Many
Eyes is the possibility of users uploading
malicious or incorrect data sets. Because there is
no guarantee that content on the site is accurate,
the burden of verifying data falls on its users. One
theme that runs through many comments is data
quality. Often users simply want clarification (are
those dollar figures adjusted for inflation?) but in
some cases they have spotted genuine errors.
One interesting result is that Many Eyes users
are finding that even official data sets are not free
from errors. A visualization of twins—triplets and
“higher-order”—births by age of mother revealed
that American women in their 50s give birth, on
average, to 9 twins! The data comes from the
National Center for Health Statistics, one of the
main governmental health statistics agencies in
the U.S. and a respectable source of information.
The incident reveals something visualization
experts already know: visual representations can
quickly expose data problems. Other problematic
data sets from official agencies have surfaced on
1 All user names have been anonymized for privacy
reasons.
Many Eyes. There may be an important role that a
public site like this can play in the debate on
information available to the public. The name of
the site comes from the open-source saying that
“many eyes make bugs shallow” and perhaps this
adage applies to data as well.
3.2 Sociability – “Harry Potter is Freaking
Popular”
One motivation behind Many Eyes was to test the
notion that visualization can be a social, playful
activity, and in several cases we have seen the
emergence of miniature games and lighthearted
interaction. Almost a month after the site
launched, user Melissa created a bubble chart2
(Fig. 3) of the fifty most popular books on
LibraryThing, a website that allows users to
catalogue their books online. The leading books
were all from the Harry Potter series, causing the
user to name the visualization Harry Potter is
Freaking Popular.
The visualizations on Many Eyes all contain a
simple pointing mechanism, where a user can
highlight a set of items for reference in a
comment. One of the authors of this paper used
this mechanism to show which books on the list
he had read. Soon after, a small game ensued
where fourteen other users highlighted the books
they had read, often with comments on their own
taste in literature. Later, the same highlighting
game was played with a visualization entitled
2006 Movies by International Box Office Gross.
The “what have I read” game required very
little from users, due to the built-in highlighting
feature. But in other cases users have gone to
considerable lengths to perform similar activities.
An example is the Countries I have Visited game.
Users create a new data set with the name of the
countries they have visited on one column and the
number of times they have been to each one of
2
Bubble charts are one of the various visualization
techniques available on Many Eyes. A bubble chart displays
a set of numeric values as circles.
these countries on the second column, and upload
that data to the site. They then create a world-map
where each country is colored based on the number
of visits.
3.3 Generating mirrors—personal and
collective
The self-revelatory games described in the
previous section are part of a broader theme: the
creation of personal “mirrors.” It is common for
people to use online journals, photo collections,
and video sharing sites to reveal aspects of their
personal lives online. Therefore, it is not entirely
surprising that a trend on Many Eyes is the
visualization of personal data.
Users have uploaded data sets that range from
collections of personal writing to running and
swimming logs to family trees. One user uploaded
and visualized a data set of monthly
measurements of his weight over the past year.
His audience can easily follow the result of a diet
that has taken him from his previous 113 kilos
(~250 pounds) to a lighter 94 kilos (207 pounds).
The user later blogged about his Many Eyes “My
Weight” graph, while also reviewing two other
Web-based tools that are geared specifically to
visually tracking one’s weight throughout the
course of a diet: skinnyr.com and traineo.com.
A different kind of Many Eyes “mirror”
depicts not a single person but an entire
community. Users have uploaded several data sets
with demographics of different virtual worlds.
There are a dozen data sets on Second Life [11]
covering everything from player gender over time
to business revenue to the distribution of active
users in different countries. Statistics on
Wikipedia [21] and World of Warcraft [22] have
also been visualized on the site.
A user on the popular community blog
Metafilter visualized social network data from that
site to Many Eyes. He then posted a short entry
about his visualization on the Metafilter
discussion board:
Figure 3: “Harry Potter is Freaking Popular” game. This sequence
shows a small series of screenshots of different users’ selections of
the books they had read.
I used IBM's Many Eyes data visualization site to create a
network map of my MetaFilter contacts and my contacts'
contacts. It got to be pretty big, so I thought I'd post it here.
This post generated a long list of posts from
other Metafilter members. Users discussed the
visualization itself, the places where each one of
them appeared in the graph, the technical
challenges of getting this kind of data, and other
possible visualization techniques. Here is a small
sample of posts:
Wow, even I'm on there. I thought I was only popular on
Metachat and in Portland. --posted by Bob
Would make for an interesting meet-up. Everyone seems to
have their seating assignments. -- posted by Harry.
I'm either less popular than I give myself credit for. Or I'm
a hydrogen molecule. -- posted by Harry
Interesting to see who the most connected people are.
Dissapointingly, I only connect to two people. -- posted by
Scott.
Here's a tag cloud from the same data. -- posted by Milton.
I feel so....small. -- posted by Mary.
Interesting that I have a lot of people linking to me yet only
a few links on the graph. Meaning, I think, that most of those
who link to me aren't in this network (monju+1 doesn't link to
them). Also, that I'm not cool. Also, that I am a failure in social
reciprocity since I don't link to most of the people who link to
me (if I had, they'd be in this network). -- posted by Ethan.
My dot is larger than I thought it would be. Go me. --
posted by Despina.
I'm trapped in an arm of the Cortex galaxy with a bunch of
other nobodys. What a rip off! -- posted by Ntrickster.
Less conspicuous communities have also
visualized themselves. NPtech, for instance, is a
group of technologists that track information on
the web that is tagged “nptech” (nonprofit
technologies such as open source software). One
of their members used our site to visualize all the
tags associated with “nptech” on del.icio.us.com.
Soon after the visualization was posted, other
Many Eyes users started asking what NPtech
stood for. Community members answered the
questions by giving a long account of the context
for the community and why it was formed in the
first place.
3.4 Sending a message
Sometimes visualizations seem to be created as
much to communicate a message as to analyze a
data set. For example, one user, “Jeremy,”
visualized the number of people with multiple
sclerosis (MS) per country worldwide. Jeremy
blogged about this visualization and the entry was
picked up a blogger friend of his, who is disabled
with multiple sclerosis. In a heartfelt post, the
blogger shows an image of the visualization and
says:
Thank you … for helping us become aware of
the fact that there are so many MSers in so many
places. We are not alone.
Visitors to the site have created visualizations
that advocate positions, that bring solace (as in the
MS example), and that illustrate issues of
importance to them. One of the distinguishing
traits in this class of visualizations is the deep
knowledge users usually have of the subject
matter they visualize. The intent seems not so
much to use visualizations as tools for exploration
and discovery as it is to make a statement and to
bond around a topic of interest. An interesting
example is a series of visualizations relating to the
Bible.
Figure 4: New Testament visualizations
Two days after launch, a user uploaded a data
set of name co-occurrences in Biblical verses and
visualized it both in a network graph and in a
treemap. The user then wrote about it on a
Christian blog that has wide readership (Fig 4).
This blog entry received more than a hundred
comments and references (many more than other
posts we examined on this blog). At first the entry
spread over a community of Christian blogs and
later reached a wider audience via such high
profile blogs as BoingBoing. In fact, a search on
technorati.com revealed that there were over 200
links back to the original blog entry (more than
twice the number of links back to the well-known
New York Times interactive infographic entitled
“Faces of the Dead” remembering 3000 fallen US
service members during the Iraq war [8]).
The Bible network diagram had a catalyzing
effect: it caused the creation of additional
visualizations of data about the Bible. One user
said on his blog that he was inspired to upload and
visualize data on the length and authorship of
New Testament documents. Another user made a
tag cloud of the first half of the Old Testament.
The trend of religious-themed visualizations
continues to this day.
Figure 5: Video blog post about Many Eyes
Two aspects to this phenomenon bear
mentioning, because they contradict some of our
own expectations for the site. First, almost all
discussion about the religious visualizations
created on Many Eyes is happening offsite. There
is a strong community of Christian bloggers that is
used to interacting through blog posts and it is no
different when they refer to the visualizations
created on Many Eyes. Some users are utilizing
the “blog this” button to add images (with links)
of the visualizations on their blogs. Other users
are more ambitious: one even created a video of
himself interacting with the original social
network graph of name co-occurrences, looking
for nodes that were not directly linked to Jesus
(Fig. 5). Given that Many Eyes has features that
encourage discussion around Many Eyes
visualizations both on and off the site [18], it is
interesting that such a high proportion of activity
occurred on external domains.
Second, we find it surprising that the most
linked visualization on the site—the graph
visualization of name co-occurrences—does not
reveal any relationships that would be unexpected
to knowledgeable Christians. Rather than showing
a discovery or novel insight, the display shows
viewers what they expect to see: Jesus dominates
the graph. He is represented as the biggest dot in
the center of the graph with the highest degree of
connection to other nodes.
The fact that the visualization reinforces a
common understanding may help viewers bond
with others of the same faith. Perhaps in this
context, a fact does not have to be new to be
worth communicating.
3.5 Innovative uses of visualizations
Many users came up with creative ways to use
visualizations. The fourteen visualization
techniques available on Many Eyes were
deliberately written to be as simple as possible.
Despite this simplicity, we found several instances
where our users managed to find unanticipated
ways to use these techniques.
One example of such as use is shown in figure
6. The data set relates to a well-known web site,
Digg.com, on which users share interesting URLS
(termed “diggs”) and annotate these URLs with
comments. The graph is a hierarchical stacked
area chart meant for displaying multiple time
series. In this case, however, the creator is
showing a “time series” with only two x-axis
points, labeled “Diggs” and “comments”. The
creator has also clicked the “percentage”
checkbox at lower right, which normalizes the
series on a percentage basis. At first this simply
looks like an error—but on close inspection, the
visualization turns out to be an interesting way of
comparing the number of diggs in a topic area to
the number of comments. It is apparent, for
example, that technology-related topics are
discussed less often relative to “world & business”
URLs. A brief conversation on the site with the
creator confirmed that this effect was intentional.
Figure 6: Novel use of Stack Graph for Categories
Several other novel uses of visualization
components emerged. One person used the bubble
chart to show data on the sizes of the planets,
yielding an elegant visual pun since the disks in
the visualization resembled planetary silhouettes.
The tag cloud was used in a variety of unexpected
ways beyond showing word frequencies (or the
contents of John’s freezer in Figure 1). One of the
most creative was a user who would take two
different, related texts (for example Apocalypse
Now and Heart of Darkness), paste them together,
and create a tag cloud of the result. As the user
explained in a comment:
This is the first in a series of what I'm going to
call a 'litmash' -- two or more texts mashed
together and visualized... In the background, I was
also thinking of William Burrough's 'cut-up'
technique, although this isn't strictly a cut-up
...more of a statistical dicing, I guest.
This example shows how the simple tools on
the Many Eyes site can be used for a complex
type of personal expression.
4 CATEGORIES OF CONTENT
The examples of section 3 provide evidence for
a rich set of activities catalyzed by Many Eyes. To
gather a broader set of data on site usage, we
collected a large selection of comments on the site
(92, representing 25% of all comments, excluding
those from members of our lab) as well as 379
data sets (24% of non-lab uploads) and a selection
of 36 external blog entries describing Many Eyes.
We then had two researchers code these
comments, data sets, and blog entries according to
preset rubrics, with discrepancies being reconciled
between the two after coding.
4.1 Comments
Our data set included 92 comments that did not
come from site developers. To classify these
comments, we relied the rubric from [6]. We
chose to reuse that scheme since it was developed
for the same purpose and so that we could
compare directly with the results of the sense.us
experiment. The rubric includes labels for
observations, questions, hypotheses, links or
references to other views, usage tips, socializing
or joking, affirmations of other comments (“Yes,
that’s right!”), to-dos for future actions (“Can
someone find inflation-adjusted data?”), tests of
system functionality, data integrity, and site
design. Note that comments are allowed to have
multiple labels.
Label Many Eyes % Sense.us %
Observation 46.3 80.6
Question 15.8 38.1
Hypothesis 11.6 35.5
Data integrity 9.5 15.7
Socializing 11.6 9
System design 11.6 9
Testing 4.2 5.6
Tips 4.2 4.1
To do 4.2 2.6
Affirmation 13.7 1.5
The results are shown in the table above. As
with the sense.us experiments, observations and
questions are the most common types of
comments, although we see fewer of these in
Many Eyes than in sense.us. The levels of
socializing and talking about the system were
about the same. The only category that is much
larger in Many Eyes is affirmations.
These results show that data analysis (in the
form of observations, questions, and hypotheses)
does occur on Many Eyes, but less intensely than
in the experiments of sense.us. One explanation
for the overall differences is that much of the
sense.us usage took place in groups of people who
were academically inclined and already knew
each other—thus more time was spent on
purposeful analysis, while on Many Eyes, where
most users are strangers, affirmations play an
important role as introductory chit-chat. Another
source of the difference may be in the type of
data: the sense.us discussions took place around a
static collection of statistics that was selected by
the experimenters to be amenable to extended
analysis, whereas discussions in Many Eyes occur
in the context of a constant flow of incoming data
sets, some of them quite simple—such as the most
popular books in a library or the top grossing
movies in a given year.
4.2 Data sets
The fact that the data on Many Eyes is
contributed by users makes comparisons with
sense.us more difficult, but raises an interesting
question as well. What sorts of data do people
want to visualize in practice? To explore this
issue, we took a sample of 379 randomly selected
data sets from the total of 1,895 on the site as of
March 2007. Two of the authors inspected these
and jointly decided on a set of labels; they then
independently coded the data sets with this rubric.
Any discrepancies were reconciled between the
two coders. The table below shows the results of
this labeling.
Label Percentage
Society (e.g. demographics) 14.0
Economics 12.7
Obscured or anonymized 12.4
Arts and culture 10.8
Web & new media 10.3
Science 10.0
Test data 9.5
Politics 7.4
Technology 6.6
Personal data (e.g. weight
loss)
6.3
Religion 5.8
Foreign language 4.7
Sports 4.2
Health 2.4
History 1.3
Education 1.0
Surveys 0.5
Environment 0.5
No label assigned 0.3
These numbers must be interpreted with
caution. Some types of data are more easily
available than others: for instance, the U.S.
government makes demographic and economic
information available for free. Furthermore, the
users of Many Eyes at this point are hardly a
random sample of web surfers due to factors
ranging from the “early-adopter” phenomenon
[10] to the type of publicity that site has
received—an article in Nature [1], for instance,
may have led to a disproportionate number of
scientists visiting the site.
Nonetheless, a few patterns are worth noting.
One is the diversity of topics. The categories are
broad but none accounted for more than 14% of
the data sets. A second point is the relatively large
sizes of categories, such as arts or religion, that
are not common topics for visualization research.
A third pattern is the large number of data sets that
we judged to be obscured in some way—e.g., a
network where the nodes were labeled with sixdigit
numbers, or schools districts referred to as
“District A,” “District B,” and so on. The presence
of these obscured data sets suggests a desire for
privacy.
4.3 Blogs
Inspired by the web-wide conversations that
occurred around the NameVoyager [20], Many
Eyes was designed to appeal to bloggers. We view
the vast net of blogs as an essential part of the
social system in which Many Eyes lives. Much of
the infrastructure of Many Eyes—from the stable
URLs that capture visualization state to the oneclick
“blog this” feature—was meant to ensure
that Many Eyes could be a full player on the Web.
Does the site succeed on these terms?
Technorati [12], a leading blog search engine,
reports that 1,215 different blogs have linked to
the site. For comparison, Technorati finds 1,605
direct links to the NameVoyager [20] and 165 to
the SmartMoney Map of the Market [19]. The
latter sites are well-known visualizations that have
been public for years, so the Many Eyes statistic
reflects a relatively high level attention among
bloggers.
Of the blog entries that mentioned Many Eyes,
we found 36 that were written by users who had
uploaded data and created their own
visualizations. Most other entries either described
the system itself or referred to visualizations that
others had created. Because we were interested in
understanding more about users’ motivations to
create visualizations, these entries provide an
excellent window into user intent. Of course, this
self-selected segment of users is unusual in many
ways, so we cannot draw conclusions about
motivations of the typical user. At the same time,
we probably can draw conclusions related to the
segment of the population that is motivated to
blog about their creations—a critical subset of the
Many Eyes audience due to its role in drawing
attention to the site.
Following the same process used to label data
sets, we categorized the motivations described by
the bloggers . The table below shows the results:
Label Number of posts
Research 4
Personal Expression 4
Journalism 12
Social interaction 3
Education 2
Analysis 10
A few of these terms merit explanation.
“Research” means performing some kind of
original investigation, as in the prose style
analysis described in section 3. Personal
expression includes the “mirror” visualizations
described above. A “journalism” blog post is one
that is designed to communicate a particular
factual message. Analysis indicates that the entry
included hypotheses about a data set.
At a high level, the blog post categorization
reflects a fairly even division between personal
topics, journalism, and analysis. To reiterate the
point made in section 3, it is interesting to contrast
this diversity of purpose with the strong
concentration on analysis that seems to drive
much of the research in the information
visualization field.
5 IMPLICATIONS FOR RESEARCH AND DESIGN
We see two primary implications of this broad
user interest in non-traditional goals for
visualization. First, some common methods of
assessing visualizations may be missing the point.
Studies that measure the time it takes users to find
outliers or make comparisons do not necessarily
predict all aspects of the value provided by
visualizations. Instead, researchers may wish to
investigate a technique’s ability to catch and hold
a viewer’s attention or to spark discussion. One
particularly important aspect of visualizations may
be the ability to catalyze conversation and
storytelling as in the collective mirrors of section
3.3. These concerns do not rule out quantitative
methods or laboratory studies—attention and
social activity are at least as amenable to
quantification as notions such as “insight” or
“discovery.” At the same time, it is also natural to
look to qualitative methods (case studies,
ethnography) as an important source of
information. Understanding non-analytic goals
may lead to the reassessment of the value of
certain aspects of visualizations—perhaps features
such as animation, which are not strictly necessary
for analytic purposes, may be seen as more
important in the context of these broader uses.
A second potential implication is an inversion
of the traditional “let the data speak for
themselves” view of statistics and visual analytics.
In fact, data visualizations may become more
broadly valuable when they can be tailored to
express a message or point of view. In other
words, when people have something to say, what
they need is a display method flexible enough to
let them say it.
Several of the examples of section 3 illustrate
this idea. The network visualization showing
Jesus at the center of the New Testament at first
seems to convey an unsurprising point. Yet this
point was seen as well worth repeating by many
of the bloggers who discussed it. In the YouTube
video that one user made of this visualization, the
Many Eyes network diagram’s support for node
rearrangement and highlighting was used to
support a miniature sermon. A second example is
the “Harry Potter” game, which shows how users
playfully exploited our highlighting mechanism to
subvert the analytical purpose of the visualization.
Designers may therefore wish to augment
communicative power of their visualizations. Here
is one example of how this might work. The
network diagram on Many Eyes has been popular,
and many people have taken advantage of the fact
that its nodes can be rearranged. Could similar
capabilities be applied to other visualizations? The
Many Eyes bubble chart, for example, follows a
traditional type of visualization design. It takes
significant CPU cycles to create an optimized
space-filling layout of disks. The algorithm works
well, and it never occurred to the designers to
allow users to change the positions of the circles.
But it might actually be a natural way to let people
categorize items or show other variables. This
change in perspective—going beyond annotation
to allow users to make wholesale modifications to
visualizations—seems potentially quite general
and fruitful.
6 CONCLUSION AND FUTURE WORK
This paper began with two main questions.
First, was the social behavior observed around
visualization in [15] [20] [6] a fluke? Second, when
lay users are given the tools to create
visualizations, what do they use the tools for?
To address the first question, consider that in
[15] [20] [6] social behavior either came
serendiptously or in limited, controlled
environments, so it was not clear such behavior
could be deliberately fostered on the open web
with a set of entirely volunteer users. The
observations of this paper indicate that we were in
fact able to stimulate a wide array of social
activity by following the design hypotheses
advanced in those early papers. Some differences
did appear between what we saw on Many Eyes
and what was reported in [6], with Many Eyes
showing less pure analytical behavior and slightly
more socializing.
While the long-term behavior on the site
remains to be seen and analyzed, these
preliminary results add some support to the idea
that data visualization can form the core of a wide
array of social behavior, both sparking
conversation and serving as a mirror or camera for
social systems. Aside from the need for longerterm
studies, there are a variety of related areas for
future work. One key issue is whether the full
scale of the internet is necessary for the
phenomena we saw. Could a large corporate
intranet support the same diversity of data
uploaded, for example? What about a single
college class?
Our observations have also provided an initial
answer to our second main question. It appears
that lay users want to exploit visualization for an
extremely diverse set of activities. A significant
amount of analysis and sensemaking did occur on
the site: this fact is supported both by particular
examples such as a graduate student performing
analysis for his thesis, and by the overall content
coding that shows many examples of
observations, hypotheses and questions. At the
same time, examples such as the “litmash” show
that completely different types of behavior occur
as well. The examples we provided are backed up
by the content coding results, which show that
many non-analytic activities are common.
Again, longer-term observation is clearly
warranted. Some of the behavior we have seen
could be due to exuberant testing of “cool”
technology, and may taper off with time.
Furthermore, in exploring both questions an
important next step is to talk with our users
directly. A natural future study would reach out to
users and openly ask about motivations for and
reactions to using Many Eyes.
Despite these caveats, it is worth
contemplating the fact that users are reinventing
visualization technology for unexpected social and
personal purposes. A historical precedent
suggests some implications: In the early days of
the telephone, businessmen formed the primary
market for telephone companies [1]. The
telephone was promoted as a replacement to the
telegraph, allowing business messages to be sent
more easily. Before the 1920s, residential sales
efforts emphasized the “business” of the
household and ways the telephone could help the
affluent household manager accomplish her tasks.
A “Telephone at Christmas” ad campaign
recommended the telephone as an aid in holiday
preparations, not as a means for giving season’s
greetings. In fact, companies discouraged clients
from socializing over the phone, reasoning that
these high-tech machines should be used for
“serious” purposes. Not until the late 1920s did
softer themes appear, linking the telephone to
sociability. Social interaction, an industry driver
today, was ignored and even resisted by the
industry in its early history.
Could a similar irony befall the field of
information visualization? Historically researchers
have emphasized data exploration, research, and
sensemaking; the current excitement around
visual analytics continues this tradition. In the
past decade, however, we have seen academic
explorations of other sorts of applications, ranging
from social visualization to ambient information
displays. While it is too early to draw firm
conclusions from user behavior on Many Eyes,
one might speculate that in the future such “nonanalytic
visualizations” will be as much of an
industry driver as traditional scientific
applications.
7 REFERENCES
[1] Robert Amar, James Eagan, and John Stasko, Low-Level
Components of Analytic Activity in Information
Visualization, In Proc. of IEEE InfoVis '05, pp. 111-117,
2005.
[2] Butler, D. Data sharing: the next generation. In Nature
446, 10-11, published online 28 February 2007.
http://www.nature.com/nature/
journal/v446/n7131/full/446010b.html, retrieved 03-30-
2007
[3] Data360: http://www.data360.org, retrieved 03-30-2007
[4] DataPlace: http://www.dataplace.org, retrieved 03-30-
2007
[5] Fisher, C. America Calling: A Social History of the
Telephone to 1940. University of California Press;
Reprint edition, 1994.
[6] Heer, J., Viégas, F.B., & Wattenberg, M. Voyagers and
Voyeurs: Supporting Asynchronous Collaborative
Information Visualization. In Proc. of CHI, 2007.
[7] Livny, M., Ramakrishnan, R., Beyer, K., Chen, G.,
Donjerkovic, D., Lawande, S., Myllymaki, J., & Wenger,
K. DEVise: Integrated Querying and Visual Exploration
of Large Datasets. In Proc. of ACM SIGMOD, May,
1997.
[8] New York Times: Faces of the Dead.
http://www.nytimes.com/
ref/us/20061228_3000FACES_TAB1.html, retrieved 03-
30-2007.
[9] Plaisant, C. The Challenge of Information Visualization
Evaluation. In Proc. of AVI 2004.
[10] Rogers, E. Diffusion of innovations. Free Press, 1995.
[11] Second Life: http://secondlife.com, retrieved 03-30-2007
[12] Spotfire DecisionSite Posters:
http://www.spotfire.com/products/ decisionsite
_posters.cfm, retrieved 03-30-2007
[13] Swivel: http://www.swivel.com, retrieved 03-30-2007
[14] Technorati: http://technorati.com, retrieved 03-30-2007
[15] Viégas, F., boyd, d., Nguyen, D., Potter, J. & Donath, J.
Digital Artifacts for Remembering and Storytelling:
PostHistory and Social Network Fragments. In Proc. of
HICSS-37, 2004.
[16] Viégas, F., & Smith, M. Newsgroup Crowds and
Authorlines: Visualizing the Activity of Individuals in
Conversational Cybersapces. Proc.of HICSS-37, 2004.
[17] Viégas, F.B. & Wattenberg, M. Communication-Minded
Visualization: A Call to Action. IBM Systems Journal.
45(4), 2006.
[18] Viégas, F.B., Wattenberg, M., van Ham, F., Kriss, J., &
McKeon, M. Many Eyes: A Site for Visualization at
Internet Scale. In Proc. of IEEE InfoVis 2007.
[19] Wattenberg. M. Visualizing the Stock Market. In Proc. of
CHI 1999.
[20] Wattenberg, M. Baby Names, Visualization, and Social
Data Analysis. In Proc of InfoVis 2005.
[21] Wikipedia: http://www.wikipedia.org, retrieved 03-30-
2007
[22] World of Warcraft: http://www.worldofwarcraft.com,
retrieved 03-30-2007