Think of walking through a physical, populous city on a summery Saturday evening. Some streets are bright and full of people, others quiet and deserted. Some cafés have boisterous crowds; in others, groups of two or three gather for serious discussion. The situations and people are detailed and vivid.
Contrast this vivid, physical city with much of the mediated world. Though tremendous amounts of social activity take place online, it feels far less vibrant. Screens of text convey little vitality. Unlike in the physical world, where people’s faces, hairstyles, clothing, and voices create memorable individual appearances, many people you encounter online create only a vague impression. A particularly outspoken or colorful writer may make a strong impact, and over time we do start to recognize and make sense of others; but in general our ability to perceive people and the subtleties of their relationships is relatively poor in online environments.
Visualization is one way to give online gatherings and conversations the vividness and legibility of the streets and cafés of the physical city. Visualizing an online conversation can help the participants better perceive each other’s roles in the community; mapping the activity in a large virtual space can provide an intuitive impression of what the different “neighborhoods” are like; and depicting an individual’s history of interaction can create a vivid portrait without necessarily revealing gender, race, or age. Visualizing social data can enliven an online space, filling it with rich detail about the inhabitants and its history, while also providing contextual cues about the tone and expectations of the setting.
Visualization is only one part of designing social interfaces—how we interact with each other via a new medium is certainly a fundamental issue—but it is an essential part of making meaning out of the vast amount of data that makes up the online world.
Social visualization is the sensory representation of social data for social purposes (Donath, Karahalios, and Viégas 1999). We call it “visualization” because the representation is usually graphical, but it can be auditory, haptic, or even olfactory. The key element is that the purpose of the visualization is social—it is meant to enhance the experience of the participants in the depicted community.
Visualizing a community can be useful to both the social scientists studying it and the people participating in it, but their goals differ significantly. The social scientists’ primary concern is to thoroughly and accurately gather and analyze the relevant statistics; they need to see the data without bias, which the neutrality of box charts and bar graphs provides. The participants’ concerns are more experiential: they want to gain an impression of their fellow participants, figure out the evolving mores of their situation, understand the social roles others are playing, and navigate the immense landscape of online social situations. Although some of the social visualizations we discuss in this book can be useful for analysis, it is not their main purpose, and the ones that feature more interpretive illustration are too subjective for social science research.
Like traditional data visualization, social visualizations highlight important patterns by turning abstract numbers and relationships into concrete images. But their use for social purposes means that they should be evocative as well as accurate: their meaning comes from the visceral impression they make, as well as from reading, say, the x- and y-axis and the labels. For example, to show which discussions on a site are active and lively, one could map the number of participants and their posting rates in several ways: line graphs, bubble charts, animations, and so on. The line chart is very clear and precise, and one can read the answer easily. But it evokes financial data or rainfall statistics more than a feeling of people and their activity. The bubble chart (whose little dots are vaguely anthropomorphic) reads more like a representation of population and presence; a deserted group makes a sparse chart, while a populous group yields a full one.
Social visualizations can be stand-alone depictions, but they are most transformative when integrated with the communication interface, such as an email program that includes visualizations of email patterns. These inhabited visualizations integrate the processes of creating the data and depicting it.
Visualization has two steps: choosing and analyzing the data and rendering it in a visible form. Both stages require special consideration for social application. Social data are often inexact or subjective: though we can accurately measure, for example, how many words someone used, whether the tone of his statement was angry or his answer helpful is more idiosyncratic. And, when visualizations represent individuals and social situations, we need to be aware of how highly attuned to subtle cues people are, and take care to avoid making misleading impressions.
The first step in any visualization is to choose meaningful data.1 For example, say you want to depict a discussion group and show the participants’ varied roles within the group. An initial thought might be to measure how often each person posts. This is an easily measured statistic, but seeing who posts prolifically and who posts rarely does not alone give much insight about people’s standing in the group. The most active participant may be a leader, but he may also be a nuisance—either a benign one, such as an inveterate writer of banal follow-ups (someone who responds to every posting with a content-free “yes!” or “me too”), or a hostile spammer, filling the conversation space with irrelevant or provocative postings. No single measure alone indicates a person’s role, but a well-chosen combination of statistics, such as the ratio of initial posts to responses plus how consistently others respond to the subject, can yield insightful patterns.
Some visualizations depict straightforward statistics: how many people are in a group, how many words are in an article, or what times of day have the most activity. But if you wish to assess the emotional tone of a message or discern the topic it addresses, you will need to further analyze the data.2 These analyses inevitably introduce some bias. For example, in visualizing a conversation, we might want to highlight changes in the emotional tone, which requires assessing whether a message conveys anger, happiness, or is simply informative. Analyzing the sentiment of the messages is qualitative, and different algorithms (or human readers) will make different interpretations. Automatic recognition of emotion is far from a solved problem, and people’s frequent use of irony and sarcasm in their everyday writing greatly complicates this task. Indeed, even for human readers, recognizing irony and humor in brief textual messages, without the facial expressions and other cues that help us face to face, is notoriously difficult. Highly processed data can reveal fascinating patterns that would be impossible to discern in the raw statistics, but the processing can introduce errors and biases. Good design needs to assess the appropriate trade-off to make for a particular use; it also should help make the choice transparent to the viewer.
The second step in visualization is to represent the data: to find expressive forms that match meaning. Visualizations can use the lines and color of traditional graphs, and can incorporate text and images; they can also use sound, motion, and interactivity to illustrate the data.3 (We will discuss these in greater depth in the next chapter, “Interfaces Make Meaning.”) The goal for the rendering might be to be accurate, legible, beautiful, evocative, or persuasive. Throughout this book, we will explore the balance among these often conflicting goals.
Visualizations can range from neutrally objective to prescriptively subjective. Neutrality is important when the visualization will be used for varied situations. A site that hosts multiple forums whose topics range from serious health problems to campy TV shows might want to use a common design for all of them. Participants in the former may be mainly concerned with expressing empathy and being able to identify trustworthy information, while those in the latter might admire witty banter, full of in-jokes and in-character writing. Here, a straightforward visualization that highlights basic social patterns, such as activity level and responsiveness, allows the viewers to see the social dynamics more clearly, without imposing a preference about which discourse style is more desirable.
For example, Loom2* is a conversation visualization that clusters messages and participants by thread (boyd et al. 2002). In designing it, we chose to show conversations as circles to reflect how people position themselves when they converse; this makes the visualization intuitive, without being prescriptive. Figure 2.1 shows a sociable and thriving group, with numerous ongoing conversations; figure 2.2 shows one where nearly all the messages are single posts, with no responses. (In this particular case, it is because the group has been overrun by spam, but there are other situations, such as announcement-only newsgroups, which similarly lack interaction yet are functioning as intended.) The visualization quickly conveys how sociable each discussion space is, while allowing the viewer to decide whether that level of sociability is desirable.
A visualization designed for a specific context can be more prescriptive. Discussion hosts, whether the owners of large sites or the writers of personal blogs, often want to set a particular tone for their community. The desired tone can vary considerably: one host might want to encourage supportive conversations and community growth, while another might want to encourage piercing critiques and rigorous arguments. Each could specify visualizations that endorse their version of desirable behavior.
Illustrative rendering makes a depiction more vivid but also more subjective. A map of discussion spaces that shows angry exchanges as dark red and purple, and supportive ones as light blue and yellow, will be easy to interpret, but it also reflects the values—and analytic techniques—of the mapmaker. We can think of these as “semantic visualizations,” conveying the designer’s interpretation of the data, not just the data itself (Donath 2002). All renderings are to some extent semantic, but typically the goal with data visualization used for science is to be as unbiased and accurate as possible, whereas semantic visualizations strive for expressiveness.
Visualizations present the patterns; as viewers, our role is to think about what they mean and why. Figures 2.3 and 2.4 show two charts from timeu.se, an interactive project that lets the viewer explore when different words and phrases appear most frequently on Twitter (Golder and Macy 2011). Figure 2.3 shows how often people mention email, and the data are unsurprising: they mention it most frequently during the weekday working hours. The plot of the word “bored” is a bit more complicated. Its lowest frequency, early morning, is predictable: people are not bored when they wake up. But its peak use is Saturday night. Are most of us really bored on what is supposed to be the most fun night of the week? Are all those parties and movies really so dull? Here we need to distinguish the question “What do people do?” from “What do people do on Twitter?” My guess is that this result reflects a combination of societal expectations and technology use. People expect to go out and have a good time on Saturday night and are likely to be disappointed and bored when they instead find themselves sitting at home. Lots of people are indeed out, and many of them are having fun—caught up in experiencing it, rather than tweeting about it. Many who are on Twitter at 8:00 or 9:00 p.m. Saturday night are at home with their computers—and they are bored.
We experience the world at multiple scales, moving fluidly among them. Standing on a high spot in a city, you see the urban landscape spread out in front of you: the neighborhoods, traffic flow, and the dance of lights at night. From this view you see overall patterns. Closer up, at street level, you notice groups of people and their interactions. You may want to know if an unfamiliar situation is safe: is that group of people just lively and boisterous, or are they aggressive and threatening? Or perhaps you are entering a party. You scan the room for familiar faces and look at the conversational groups, seeking one to join. Some are clearly talking about something light and funny, while others appear to be discussing more somber issues. One group of people is talking intimately, leaving no room for a newcomer, while another seems welcoming, the people making eye contact with you and shifting to make room. At a closer scale, you focus on an individual, perhaps a friend you are chatting with or a stranger who has caught your attention, and you study how he is dressed, his mannerisms, and so on, in order to understand more about who he is and what he is thinking.
In filmmaking, this multiscale perspective translates into the use of long shots, medium shots, and close-ups. Long shots, often called “establishing shots,” show a wide perspective of the scene, giving the viewer a sense of the overall setting. Medium shots show a few people in their frame, allowing the viewer to see the relationships among them and follow the dynamics of a conversation. Close-ups portray an individual.
We can think of social visualizations in terms of these scales. One might want to depict an entire community, an ongoing conversation, or a specific individual. An overview of a community, the long shot, helps you find your way through a large and complex social landscape. This distant view is for observing patterns; users interact with it by manipulating the representation for their personal understanding. The medium shot, such as a visualization of a conversation, helps you make sense of relationships and roles. Here the viewer may be an active participant in the discussion. The close-up is a data portrait—a way of seeing and recognizing an individual based on her history and reputation, rather than her facial features.
In the physical world, moving between these scales is a matter of shifting focus, and the transition is typically smooth and continuous. Online, however, different interfaces handle different scales. Finding a way to smoothly transition between these scales online is a difficult challenge.4 Zoomable interfaces allow you to go from a distant view to a close-up, but your role remains that of an observer (see figures 2.5 and 2.6). Designing an interface that smoothly shifts from observation to participation at the appropriate scales remains an open design challenge.
In Sophie and Bruno Concluded, Lewis Carroll wrote about the futility of a truly accurate map:
“That’s another thing we’ve learned from your Nation,” said Mein Herr, “map-making. But we’ve carried it much further than you. What do you consider the largest map that would be really useful?”
“About six inches to the mile.”
“Only six inches!” exclaimed Mein Herr. “We very soon got to six yards to the mile. Then we tried a hundred yards to the mile. And then came the grandest idea of all! We actually made a map of the country, on the scale of a mile to the mile!”
“Have you used it much?” I enquired.
“It has never been spread out, yet,” said Mein Herr: “the farmers objected: they said it would cover the whole country, and shut out the sunlight! So we now use the country itself, as its own map, and I assure you it does nearly as well.” (Carroll 1893, 169)
A similar theme was explored by Jorge Luis Borges (1998) in a very short story entitled “On Exactitude in Science.”
Abstracting and simplifying, not absolute verisimilitude, provide the value of the map.
In the rest of this chapter, we will look at social landscapes, the equivalent of the establishing shot for online social spaces. Medium shots, focused on the interactions within a group, are featured in chapter 6, “Visible Conversations”; close-up individual portrayals are the topic of chapter 8, “Data Portraits.”
As we have seen with Visual Who, often the most challenging part of depicting an online community is mapping its “landscape.” Visual Who used the network of common interests as manifest in their mailing list subscriptions to create its interactive community map. This is but one of many possible approaches to mapping a community.
Another approach is a traditional geographic map, which places people and events in their physical locations; it is useful when those locations are especially meaningful (see figure 2.7). Often, however, data-centric mappings reveal more interesting patterns. Freed from the constraint of geographic location, such maps of social landscapes can use spatial features such as size and adjacency to convey relationships within a community.
The mutability of the online world presents a further challenge. Our familiar, physical surroundings are deeply rooted in the past; geological formations and historical events shape where roads are laid down and cities grow. Maps of physical space are anchored to this geography. Maps of virtual spaces, however, have no such firmament; it is a space with no geography.
Traditional maps are simplifications of geographic reality, highlighting different aspects of the terrain—streets, history, population, geology—depending on the purpose of the map. Their function is to abstract an orderly and comprehensible view from the immense and messy complexity of reality. Similarly, the makers of maps of virtual spaces must abstract key data from the messy complexity of vast archives.5
Here we will look at two approaches for mapping a large-scale virtual space, the handcrafted map and the algorithmic visualization. The first example, Updated Map of Online Communities by Randall Munroe, is a hand-drawn map of major online communities as of 2010. A handcrafted map can draw upon the artist’s knowledge, humor, and opinions to create a witty and insightful depiction. However, it is difficult to generalize and must be (painstakingly) redone for new communities or even as an already-mapped one evolves. Often, hand-drawn maps are subjective; the artist’s beliefs shape its world. This is an excellent quality when the goal is to editorialize or critique, but viewers seeking to observe the actual social patterns of a community need a more objective view.
The second, Netscan by Marc Smith, is an algorithmically generated map of discussions on Usenet circa 2004. Algorithmic maps are more generic, and often more objective, though the choice of data and display inevitably introduces some subjectivity—and possibly a lot (Gillespie 2012, 2013).6 While an algorithmic map cannot match the idiosyncratic wit of the hand-drawn one, it can, once programmed, depict numerous communities and update them frequently. Algorithmic maps can incorporate immense amounts of data and show us interesting and sometimes surprising and anomalous patterns.
Our focus in this book is on designing computer generated depictions; the goal in looking at hand-drawn maps is to see what we can learn for that endeavor from the construction of a handmade one.
Updated Map of Online Communities by Randall Munroe, the originator of the online cartoon xkcd, is a hand-drawn map of the major online communities as of 2010 (see figure 2.8). Sites are grouped thematically. For example, the mainland is made of social network sites, dominated by Facebook, and surrounded by geographical features such as the “Plains of Awkwardly Public Family Interactions” and the “Northern Wasteland of Unread Updates,” while various technologies for real-time conversation—Google Talk, Skype, and ICQ—are islands off the “Sea of Protocol Confusion.” The map is enjoyable to peruse because of the meaningful juxtapositions and distributions of its landmarks. Its insight, humor, and irony will be beyond the capabilities of computer-generated maps for quite some time.
Though the layout is witty, it is not entirely fanciful. Geographic size represents “total social activity in a community—that is, how much talking, playing, sharing, or other socializing happens there.” For readers familiar with the online social sites of that era, the map is intuitively accurate. The designer, Randall Munroe, went beyond straightforward statistics in his assessment of “social activity.” He points out that with the rapid rise and subsequent abandonment of social sites, membership numbers are not the most meaningful: one wants to measure vitality, a more salient, but also subjective, quantity. “Estimates are based on the best numbers I could find, but involved a great deal of guesswork, statistical inference, random sampling, nonrandom sampling, a 200,000-cell spreadsheet, emailing, cajoling, tea-leaf reading, goat sacrifices, and gut instinct (i.e. making things up).”
A great deal of knowledge goes into making a map such as this one, beyond the multivariate assessment of “vitality.” Munroe grouped communities by their technological underpinnings, their role in society (corporate communication versus game playing versus subversive meme generation), their national origin, and more. The layout makes no claim to satisfying all these criteria equally, and, indeed, there is no two-dimensional layout that could. It is, however, a meaningful layout, in which each adjacency depicts an interesting relationship. A key lesson from this is that layout is important, but imperfection is tolerable. There are inevitably other connections that, in any particular layout, are not shown. Each possible layout would yield different insights and comparisons.
The Updated Map of Online Communities presupposes that the viewer has considerable knowledge of these social spaces. For the uninitiated, “Duckface Mountains” and “Bieber Bay” are unremarkable, indeed meaningless, but for those in the know, they are little jewels of humor.7 Much of the pleasure in viewing this map is in finding familiar sites and seeing how they have been depicted.
Perhaps the most important lesson to be learned from this map is that visualizations should be interesting to peruse. Monroe’s map is both informative and funny. It includes many sites that the typical xkcd reader, though quite Internet savvy, may be unfamiliar with, such as Chinese and other non-Western social network sites. Its details reward careful viewing. For example, Facebook has been criticized for making it difficult for users to adjust their privacy settings, and in the Facebook subcontinent, beyond the foothills of “People You Can’t Unfriend,” is a small lava pool, with the island of privacy controls set unreachably in the middle.
Algorithmically generated maps of online communities take data about a community and render this information visually. The designers of such maps do not draw each line and detail; their craft is instead in choosing interesting data, creating algorithms that reveal key patterns, and writing programs that render them clearly and engagingly.
Though we cannot computationally generate incisive wit, as in Munroe’s Online Communities map, there are other ways to intrigue the viewer. We can add information, giving the viewer a glimpse of what each space is like; a map of a conversation space can allow the viewer to zoom into the actual text of a particular discussion. More generally, interactivity—letting the viewer explore the space, such as by varying the layout or delving more deeply into particular regions—helps people make sense of vast datasets and complex material, showing further detail or other dimensions of it. It encourages the viewer to spend more time examining the data, trying different combinations and seeing the relations that emerge (Donath 1995; Keim 2001; Shneiderman 1994; Yi et al. 2007).
Tree graphs (see figure 2.9) have traditionally been used to depict hierarchical data, but the graphs do not show comparative measurements such as size. Treemaps are a technique for depicting such data using nested rectangles (Balzer, Deussen, and Lewerentz 2005; Bederson, Shneiderman, and Wattenberg 2002; Shneiderman 1998–2009). The advantage of the treemap is that it can show very large datasets in a limited amount of space and make it possible to compare data about the different branches of the hierarchy at a quick glance. Marc Smith and colleagues’ treemap visualization of Usenet is an example of this kind of algorithmic map (see figures 2.10, 2.11, and 2.12).
Usenet was a very large discussion site with thousands of hierarchically organized topical newsgroups (see chapter 7, “Contested Boundaries,” for more on Usenet history). At the top level were divisions such as rec (for recreation), comp (for computers), sci (for science), and alt (alternative, things that did not fit elsewhere). “Comp” was then further subdivided into areas such as lang, comp.graphics*, and comp.theory*. (The asterisk is a placeholder for all the groups at the next lower level.) Specific groups would then be comp.theory.cell-automata and comp.theory.self-org-sys. Other hierarchies had groups such as rec.sport.soccer, rec.pets.cats, or alt.sex.bondage. The different discussion groups had distinct and varied personalities. Even ostensibly similar ones, such as two computer language groups, could be quite disparate: one might be rigorous and technical, while the other mixed social chitchat in with answers to beginners’ questions. Some groups were quite civil, while others were plagued with hostile comments. (In chapter 6, “Visible Conversations,” we will look at some approaches for depicting these differences.)
The major categories were also distinctive: comp.* and sci.* were home to the more serious research and technical discussions that were the original foundation of Usenet; soc.* and rec.* contained discussions and arguments ranging from politics and religion to the care of houseplants. Most of Usenet was governed by a set of strictly administered rules about who could add new discussion groups and on what topics. In contrast, in the alt.* hierarchy, which was created to provide for everything that did not fit the main groups, anyone could start a new group on any topic. Alt.* included such groups as alt.suicide.holiday, alt.tasteless, and alt.usenet.kooks.
To make a treemap from a hierarchy, the overall space is first divided up into smaller rectangles, one for each of the top levels of the hierarchy. The size of the rectangles is in proportion to some quality of the category. In figures 2.10 and 2.11, it is the number of postings; in figure 2.12, it is the number of replies. Then, each rectangle is similarly divided into smaller rectangles for each of its subcategories, and so on for as many subdivisions as there are levels of hierarchy. Thus, in figure 2.11, alt is the large top-level category on the left side, taking up more than half the main rectangle. Alt.binaries, outlined in yellow, is a big subcategory, which in turn includes individual groups such as alt.binaries.misc, and further subcategories such as alt.binaries.mp3, which includes alt.binaries.mp3.sounds.
Usenet was changing during the times these maps depict. In the 1990s, it was the major site for online conversation. By 2010, the xkcd Updated Map of Online Communities would show Usenet as a tiny island, labeled “Usenet, still there!” How do these maps tell that story?
At first glance, there seem to be no signs of Usenet’s deterioration. In fact, it looks to be thriving. Green means growth, and the 2004 map is much greener than the 2000 one. But a closer look shows more troubling developments. In figures 2.10 and 2.11, the categories alt.binaries.* and rec.* have been outlined in yellow. Binaries are data files, meant to be read by machines, not humans. The binaries that are posted to Usenet are often illicit: pornography and pirated music, games, and movies. And it is alt.binaries, the non-conversational repository of this material, that shows the most vigorous growth in number of postings. Rec.* and the other original top-level hierarchies—once the site of lively conversations, exchanges of technical information, and social support—have become comparatively small. By 2004, over half of all postings to Usenet were in the alt.* hierarchy, and the majority of them were binaries. Although these maps cannot tell us why it happened, they do show that while the quantity of data uploaded to Usenet rose, its quality—at least in terms of sociability and human interaction—declined precipitously.
The question one wants answered determines which data are best to map. A count of the number of postings answers “Where is there the most activity?” It does not tell you what kind of activity it is, whether it is friendly discussion, political disagreements, blatant advertising, or illegal uploads. It does not tell you where the most resources are being used, that is, what group is using the most storage and bandwidth. But it does give you a general idea of the trends in what people are doing on the site: the maps in figures 2.10 and 2.11 show a Usenet in which people are uploading binaries more and talking about compilers less. Similarly, a street map, though a useful thing in its own right, provides only subtle hints about, say, the terrain or a neighborhood’s economic status.
The person seeking a congenial conversation needs a different map, created from different data. Smith made another Usenet treemap, this one based on the number of replies, rather than all postings (see figure 2.12). Since spam and binary uploads tend to be stand-alone posts, this visualization provided a better indication of where social interaction was still occurring; in this map, alt.binaries appears relatively small, as few of its immense number of posts are part of a conversation.
Reply counts alone are insufficient to guide the viewer to a vibrant conversational space. A group might have many replies, but still many times more junk postings, which people do not want to be forced to sift through to find something useful. For this goal, mapping relationships such as the ratio of replies to postings or the average number of exchanges in a conversational thread would create a more informative image. Smith and colleagues used color to show whether a group was growing or shrinking, but one could create similar maps to depict almost anything: the size of the postings, the number of different posters, the predominant language, and so on. A version of this map that allowed viewers to experiment with looking at different statistics would be fascinating to explore.
Algorithmic maps do not have the overt idiosyncrasy and editorial commentary that a hand-drawn map may feature, but they are still subjective: the designer’s choices of data and depiction shape the message they convey. Sometimes the map’s intent is to be objective, in which case its subjectivity is, like the distortions inherent to physical maps, inevitable yet regrettable. Sometimes, though, the exaggerations and highlights are deliberate: a map’s purpose might be political, to persuade or take a position. Much as newspapers distinguish “news,” which aims for objectivity, from “editorial,” which is clearly labeled as subjective, a map should ideally make its purpose and biases clear to the viewer.8
The big question we are addressing here is how to navigate an immense virtual social space. Today, this question arises when one looks at Twitter and tries to decide whom to follow, out of the possible millions. It arises when looking at a newspaper such as the New York Times online, where many articles generate hundreds of comments. It arises when visiting any of the innumerable sites that are, like Usenet was, home to myriad conversational communities centered on diverse topics and with greatly varying levels of sociability. Even seemingly niche sites, such as “Digital Photography Review” (http://www.dpreview.com), host numerous forums, with some, such as “Canon EOS 7D/60D–10D Talk,” comprising millions of messages on hundreds of thousands of topics.
Maps are an aid to navigation, an invitation to explore. The goal of mapping the social landscape is to help the viewer understand what is possible and where one might go—where to find a compatible community, a lively debate, an up-to-date source of news or gossip. Without a map, we can search to find a discussion of a particular topic, or follow the recommendations of writers we already read, but these approaches can be narrow. A map designed to highlight relevant social patterns helps the viewer find unfamiliar yet promising places.
A map that shows activity helps us see the vibrancy of the online world: it makes the virtual crowd come alive. Though many of the lively areas may not be relevant for you (e.g., fan sites for celebrities you don’t care about, discussions about products you don’t own or diseases you don’t have), it is interesting to know where and what they are, to be aware of what fascinates your fellow participants.
Even without the set geography of the physical world, the map of a virtual space can still reflect its structure. A treemap suits a site with a fundamentally hierarchical structure, such as Usenet or a newspaper with topical sections. Other structures, such as networks, call for different depictions. The “Interactive Persian Blogosphere Map” depicts a network of blogs (see figure 2.15). This map also combines algorithmic and hand-drawn techniques: each dot is a blog, and an automated program places them algorithmically based on the interconnections between the blogs (their links to each other) and their common external links. The map designers labeled the clusters by hand and annotated both the clusters and the circled blogs with descriptions of their contents, putting them into the larger context of Iranian politics (Kelly and Etling 2008). Using a computer to make the algorithmic layout allowed them to locate a large number of blogs using fairly complex criteria, while the hand annotation adds a human interpreter’s insight to the map. Visual Who, the project that opened this book, is another way to map a network; we shall look at others in chapter 4, “Mapping Social Networks.”
There are multiple ways to depict a single site. The treemaps we have been examining were part of a larger endeavor that explored different ways of mapping Usenet; a related map (see figure 2.16) showed the newsgroups as a network, where the number of cross-postings (postings addressed to two or more groups at the same time) was the measure of tie strength. The treemap and the network depiction are both “right”—they highlight different patterns.
The visualization’s sensory quality is also important. Voronoi treemaps (figure 2.17), for example, present the same information as traditional treemaps, but are gracefully organic, appearing to have evolved naturally, rather than as if they were mechanically sliced (Balzer, Deussen, and Lewerentz 2005). One might prefer to use this approach for depicting a discussion space, because it gives the impression of a living, active community. In the next chapter, “Interfaces Make Meaning,” we discuss the elements that contribute to the sensory qualities of online interfaces and how these, as well as metaphor and interactivity, produce the feel of an online experience.
A history of eighteenth-century English road infrastructure makes an interesting point about the social change from wayfaring to map-reading (Guldi 2012). In earlier days, when maps were vague sketches, travelers needed to ask for assistance throughout their journey: what to do at the next turn, where to find a decent inn, were the roads passable? As infrastructure became more reliable and maps became more accurate (the detailed surveying that was needed to build the road also greatly improved mapmaking ability), people became more self-reliant in their travels. They no longer needed to engage their fellow travelers for information about the next turn to take or cultivate good will to get reliable news of the conditions. When we travel today, most of us rely on GPS, guidebooks, and maps, and as a result travel can be an isolated experience. It is only when we get lost or if there is an unexpected problem that we need to stop to talk with other people. Maps let us navigate asocially.
A parallel transformation happened in the early days of the Web. Before search engines existed, people relied on each other to find their way around the burgeoning set of Web pages. Users created home pages that, in addition to describing themselves and their interests, included a list of links to things they had found to be useful or funny on the Web. The most comprehensive or well-curated lists of recommended pages became themselves among the Web’s most popular sites. Then search engines were developed. Today information seeking on the Web no longer requires that you find the individual people who create the best lists and guideposts; now algorithms compile this information for you.9 You simply type a query.
It is easy to romanticize the wayfaring travel of the past. Being dependent on the kindness of not always well-meaning strangers can be uncomfortable, as well as dangerous. Relatively few people traveled then, and not for recreation. The advent of easily followed maps helped fuel an immense leisure tourism industry (Strickland 1998). Online social way-finding does not raise the specter of being robbed on a deserted country road. But relying on social connections to find information is far less efficient than performing a quick search of a fully indexed and massive trove of data. Few people would argue for a return to the days when web users relied on Yahoo, “Cool Site of the Day,” “Justin’s Links from the Underground,” or other handcrafted lists to find interesting material. But in other cases, something valuable has been lost in the achievement of efficiency.
At its height, Usenet hosted thousands of active conversations and was a place people went to find stimulating discussions, social support, and, above all, useful information. Obtaining this information was a social process: you were expected to familiarize yourself with the past discussions and social mores, to be a part of the community, not simply an opportunistic answer-seeker. In the mid-1990s, Usenet’s data was imported to the Web and it became much easier to search it (Brown 1999; Hauben 2002). Information seeking became divorced from relationship building, contributing to the collapse of community on this site.
Maps of communities allow us to see histories and relationships that would otherwise require extensive study to discern. As social visualizations, their purpose is to aid in social engagement. But do they have the paradoxical effect of making us less social, less dependent on building relationships to learn about the surrounding milieu? I would argue no, that they ultimately aid sociability; they make it easier for us to make initial contacts, to find the person we want to speak with, and to understand social nuances that might otherwise escape us. But we need to be cognizant of this tension, and use design to foster sociability, not supplant it.