Fernanda Viégas, Martin Wattenberg
This is an exciting moment for visualization. It’s a time when the mainstream media is embracing sophisticated techniques born in university research labsa time when you can open The New York Times and see complex treemaps and network diagrams. But just as exciting is the fact that some new visualizations, ones that get people talking and thinking about data in a new way, are emerging from outside the academy as well.
This is starting to happen often enough that it’s worth coining a term for techniques that originate outside the research community. Borrowing terminology from the design world, we’ll call them “vernacular” visualizationsin a nod to Tibor Kalman’s admiration of “low” art . This article focuses on one ubiquitous type of streetwise visualization: tag clouds. Born outside the world of computers, they were raised to maturity by Web 2.0 sites coping with an unwieldy world of collective activity. Tag clouds are an eclectic bunch spanning a variety of data inputs and usage patterns that defy much of the orthodox wisdom about how visualizations ought to work.
Tag Clouds: A History
The basic look of a tag clouda combination of many different type sizes in a single viewgoes back at least 90 years to Soviet Constructivism. Beyond the surface style, however, a tag cloud usually has a particular purpose: to present a visual overview of a collection of text. By this criterion, the first example may have been the outcome of an experiment carried out by social psychologist Stanley Milgram in 1976 . Milgram asked people to name landmarks in Paris, and then created a collective “mental map” of the city using font size to show how often each place was mentioned (Figure 1).
Almost 20 years later, similar diagrams were created by a computerbut a fictional one. In Microserfs, Douglas Coupland’s 1995 novel, one of the characters writes a program to pick out random phrases from his electronic diary; the resulting “subconscious files” are reproduced in the book.
In 1997 real life caught up with Milgram’s collective unconscious and Coupland’s fictional computer. That year programmer Jim Flanagan wanted a way to show which search terms had led people to his website. Varying type sizes was easy; in a page defined by HTML, he created a simple Perl script“Search Referral Zeitgeist”and the resulting graphic was eyecatching. However Flanagan’s script remained an obscure curiosity.
By 2001 clouds of words worked their way into mainstream media when Fortune magazine brought the “cloud aesthetic” to the world of finance . A piece entitled “Money Makes the World Go ‘Round” mapped the corporate landscape with circular masses of text showing the 500 largest corporations in the world, each cloud representing companies in a particular country (Figure 2). According to the graphic’s creator, John Tomanio, this was an independent invention. Perhaps something was in the air!
Then, in 2002, a young photo-sharing site called Flickr needed a way to show how users had classified, or “tagged,” their images. Borrowing Flanagan’s idea, Flickr added a “tag cloud” that showed the popularity of various tags using font size. The resulting graphic (Figure 3), which had a distinctive jumbled quality and whose most prominent words included “wedding,” “beach,” and “Nikon,” provided an instant overview of the site’s pictures and gave this kind of graphic a name.
Other tag-based sites, such as del.icio.us and Technorati, followed Flickr’s lead and the technique soon became a hallmark of Web 2.0 design. These tag clouds function as aggregators of activity being carried out by thousands of users, summarizing the action that happens beneath the surface of socially oriented websites. Two years after Flickr’s tag cloud appeared, John Maedarenowned graphic designer and computer scientist at the MIT Media Labdeclared this type of graphic “The Greatest Diagram of 2004” . Today it seems like any site with tags exhibits its own flavor of visualization.
Aggregation of tags, however, is not the only use of tag clouds. There have been several attempts at using this technique for analytical tasks as well. These usually boil down to the examination of textual documents ranging from political speeches (Figure 4) and collections of blog posts to academic articles and entire books. Some of these efforts incorporate time, allowing users to quickly examine a collection of sequential documents (Figure 5). In such cases it might be more logical to refer to the technique as a “word cloud” since it shows word frequency in a text passage instead of tags on a site. (In a sense this is returning to the roots of Milgram’s and Coupland’s diagrams.)
Adding steam to the analytical power of tag clouds, a few sites have created alternative versions of the technique. Our own IBM Research site Many Eyes, which lets users upload and visualize data in a variety of ways, features two flavors of tag clouds: the traditional one-word and an unconventional two-word view. The two-word cloud shows the most frequent two-word phrases in a body of text (disregarding stop words). This simple mechanism gives viewers a much different perspective on the text being visualized, getting closer to some of the themes in the content. For example, the one-word and two-word clouds of a political speech uploaded to Many Eyes give viewers a dramatically distinct understanding of the material (see Figures 6a and 6b).
An Information-Visualization Technique
From a traditional information-design perspective, the rapid adoption of tag clouds is surprising because the method has glaring theoretical problems. Long words get undue emphasis over short ones, it is cumbersome to find any single word, and font sizes can be difficult to compare. Alphabetical ordering means that sets of related words are scattered: “East” is close to “Easter” but far from “west.”
Recent empirical work supports these theoretical concerns. In one experiment subjects shown tag clouds performed worse in word recognition and overall sense making compared with people looking at simple vertical lists of words in alphabetical order . The alphabetical ordering in tag clouds is an important organizing principle because it provides the only way to visually search for specific items in the display. Yet another study has determined that some users are oblivious to the fact that words are alphabetically organized . So there’s a puzzle: If tag clouds don’t provide quantifiable benefits and if people are unaware of how items are organized in the visualizations, how and why are tag clouds being used?
Hearst and Rosner suggested one possible answer when they noted that tag clouds seem to serve another purpose, as social signifiers that imply a friendly atmosphere and provide a point of entry into a complex site . Web 2.0 sites tend to attract thousands, sometimes millions, of users who contribute contentgrasping the scale and diversity of these contributions is a challenge. Therefore, having tag clouds that summarize some of this activity in a simple manner can be a valuable asset for the community of users. In a sense, these clouds may act as individual and group mirrors which are fun rather than serious and businesslike. Indeed, this all-word visualization is a diagram that even a mathophobe can love.
But what about tag clouds that appear outside the realm of social tagging? Our experience on Many Eyes suggests a few other uses. In some cases, tag clouds function as portraits of individuals rather than groups. One person uploaded the text from 20 blogs he read (10 from men, 10 from women) to create a gallery of verbal snapshots. Each cloud was accompanied by commentary on what it revealed about the blogger’s personality. While this certainly was a form of analysis, the engaging nature of tag clouds made the technique a natural fit.
In other cases, however, tag clouds seem to be used for more traditional analytical purposes. Numerous bloggers have written about the tag clouds of political speeches and have painstakingly examined the differences among politicians. Here users actively find, analyze, and communicate patterns in text, rather than merely obtaining a glimpse of the “gist” of a piece of work. Despite the theoretical concerns, tag clouds have become a tool of choice for analysis.
A tag cloud is truly a “vernacular” techniqueone that does not come from the visualization community, and that violates some of the golden rules of traditional visualization design. Nevertheless, the tag cloud’s widespread popularity and flexibilityplaying a starring role in situations ranging from psychological experiments to fiction writing to political analysissuggest that it passes the test of applicability. One might say that tag clouds work in practice, but not in theory.
This failure of conventional wisdom deserves attention because it points to new possibilities. The increasing demand for tag clouds indicates that there is an important class of data that users want to visualize: unstructured text. In addition, the value users draw from such visualizations, as social signalers or as tools for collective text analysis, suggests that experts in information design might want to rethink the purpose and goals of their creations. In this moment when nonacademic designers are adopting academic visualization techniques, theorists can return the favor and take inspiration from the current burst of creativity in vernacular visualization.
The authors would like to thank Marti Hearst for insightful comments on an early draft of this article, and Jonathan Grudin for the invitation to contribute to the Timelines forum.
Many Eyes: http://many-eyes.com
Tagline Generator: http://chir.ag/tech/download/tagline
Fernanda B. Viégas and Martin Wattenberg are research scientists at IBM’s Visual Communication Lab. Viégas is known for her pioneering work on depicting chat histories and email. Wattenberg’s visualizations of the stock market and baby names are considered Internet classics. Both Viégas and Wattenberg are also known for their visualization-based artwork, which has been exhibited in venues such as the Museum of Modern Art in New York, the London Institute of Contemporary Arts, and the Whitney Museum of American Art. The two became a team in 2003 when they decided to visualize Wikipedia, leading to the “history flow” project that revealed the self-healing nature of the online encyclopedia. Their current project, Many Eyes, explores the power of Web-based visualization and the social forms of data analysis it enables.
Figure 6B. Two-word tag cloud of the same
Barack Obama speech. This version reveals three main themes in
the text: “health care,” “special interests,” and sending the
“troops home.” Created on Many Eyes.
©2008 ACM 1072-5220/08/0700 $5.00
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2008 ACM, Inc.