archive.Joe Lamantia.com

In Pivoting on tags to create better navigation UI Matt McAllister offers the idea that we’re seeing “a new user interface evolving out of tag data,” and uses Wikio as an example. For context, he places tag clouds within a continuum of the evolution of web navigation, from list views to the new tag-based navigation emerging now.

It’s an insightful post, and it allows me to build on strong groundwork to talk more about why and how tag clouds differ from earlier forms of navigation, and will become [part of] a new user interface.
Matt identifies five ‘leaps’ in web navigation interfaces that I’ll summarize:

A Lesson in ‘Listory’
As Matt mentions, all four predecessors to tag based navigation are really variations on the underlying form of the list. There’s useful history in the evolution of lists as web navigation tools. Early lists used for navigation were static, chosen by a site owner, ordered, and flat: recall the lists of favorite sites that appeared at the bottom of so many early personal home pages.

These basic navigation lists evolved a variety of ordering schemes, (alphabetical, numeric), began to incorporate hierarchy (shown as sub-menus in navigation systems, or as indenting in the left-column Matt mentions), and allowed users to change their ordering, for example by sorting on a variety of fields or columns in search results.

From static lists whose contents do not change rapidly and reflect a single point of view, the lists employed for web navigation and search results then became dynamic, personalized, and reflective of multiple points of view. Amazon and other e-commerce destinations offered recently viewed items (yours or others), things most requested, sets bounded by date (published last year), sets driven by varying parameters (related articles), and lists determined by the navigation choices of others who followed similar paths.)

But they remained fundamentally lists. They itemized or enumerated choices of one kind or another, indicated implicit or explicit precedence through ordering or the absence of ordering, and were designed for linear interaction patterns: start at the beginning (or the end, if you preferred an alternative perspective – I still habitually read magazines from back to front…) and work your way through.

Tag clouds are different from lists, often by contents and presentation, and more importantly by basic assumption about the kind of interaction they encourage. On tag-based navigation Matt says, “This is a new layer that preempts the search box in a way. The visual representation of it is a tag cloud, but the interaction is more like a pivot.” Matt’s mention of the interaction hits on an important aspect that’s key to understanding the differences between clouds and lists: clouds are not linear, and are not designed for linear consumption in the fashion of lists.

I’m not saying that no one will read clouds left to right (with Roman alphabets), or right to left if they’re in Hebrew, or in any other way. I’m saying that tag clouds are not meant for ‘reading’ in the same way that lists are. As they’re commonly visualized today, clouds support multiple entry points using visual differentiators such as color and size.

Starting in the middle of a list and wandering around just increases the amount of visual and cognitive work involved, since you need to remember where you started to complete your survey. Starting in the “middle” of a tag cloud – if there is such a location – with a brightly colored and big juicy visual morsel is *exactly* what you’re supposed to do. It could save you quite a lot of time and effort, if the cloud is well designed and properly rendered.

Kunal Anand created a visualization of the intersections of his del.icio.us tags that shows the differences between a cloud and a list nicely. This is at heart a picture, and accordingly you can start looking at it anywhere / anyway you prefer.

Visualizing My Del.icio.us Tags

We all know what a list looks like…

iTunes Play Lists

What’s In a Name?
Describing a tag cloud as a weighted list (I did until I’d thought about it further) misses this important qualitative difference, and reflects our early stages of understanding of tag clouds. The term “weighted list” is a list-centered view of tag clouds that comes from the preceding frame of reference. It’s akin to describing a computer as an “arithmetic engine”, or the printing press as “movable type”.

[Aside: The label for tag clouds will probably change, as we develop concepts and language to frame new the user experiences and information environments that include clouds. For example, the language Matt uses – the word ‘pivot’ when he talks about the experience of navigating via the tag cloud in Wikio, not the word ‘follow’ which is a default for describing navigation – in the posting and his screencast reflects a possible shift in framing.]

A Camera Obscura For the Semantic Landscape
I’ve come to think of a tag cloud as something like the image produced by a camera obscura.
Camera Obscura

Where the camera obscura renders a real-world landscape, a tag cloud shows a semantic landscape like those created by Amber Frid-Jimenez at MIT.

Semantic Landscape

Semantic Landscape

Like a camera obscura image, a tag cloud is a filtered visualization of a multidimensional world. Unlike a camera obscura image, a tag cloud allows movement within the landscape. And unlike a list, tag clouds can and do show relationships more complex than one-dimensional linearity (experienced as precedence). This ability to show more than one dimension allows clouds to reflect the structure of the environment they visualize, as well as the contents of that environment. This frees tag clouds from the limitation of simply itemizing or enumerating the contents of a set, which is the fundamental achievement of a list.

Earlier, I shared some observations on the structural evolution – from static and flat to hierarchical and dynamic – of the lists used as web navigation mechanisms. As I’ve ventured elsewhere, we may see a similar evolution in tag clouds.

It is already clear that we’re witnessing evolution in the presentation of tag clouds in step with their greater visualizatin capabilities. Clouds now rely on an expanding variety of visual cues to show an increasingly detailed view of the underlying semantic landscape: proximity, depth, brightness, intensity, color of item, color of field around item. I expect clouds will develop other cues to help depict the many connections (permanent or temporary) linking the labels in a tag cloud. It’s possible that tag clouds will offer a user experience similar to some of the ontology management tools available now.

Is this “a new user interface”? That depends on how you define new. In Shaping Things, author and futurist Bruce Sterling suggests, “the future composts the past” – meaning that new elements are subsumed into the accumulation of layers past and present. In the context of navigation systems and tag clouds, that implies that we’ll see mixtures of lists from the four previous stages of navigation interface, and clouds from the latest leap; a fusion of old and new.

Examples of this composting abound, from 30daytags.com to Wikio that Matt McAllister examined.

30DayTags.com Tag Clouds

Wikio Tag Cloud

As lists encouraged linear interactions as a result of their structure, it’s possible that new user interfaces relying on tag clouds will encourage different kinds of seeking or finding behaviors within information experiences. In “The endangered joy of serendipity” William McKeen bemoans the decrease of serendipity as a result of precisely directed and targeted media, searching, and interactions. Tag clouds – by offering many connections and multiple entry paths simultaneously – may help rejuvenate serendipity in danger in a world of closely focused lists.

Though you may not have noticed it at first (I didn’t – they’re located a few steps off the front page), the recently launched design of NYTimes.com includes tag clouds. After a quick review, I think their version is a good example of a cloud that offers some increased capabilities and contextual information that together fall in line with the likely directions of tag cloud evolution we’ve considered before.

Specifically, the New York Times tag cloud:

NYtimes.com Tag Cloud

The NYTimes.com tag cloud shows the most popular search terms used by readers within three time frames: the last 24 hours, the last 7 days, and the last 30 days. Choosing search terms as the makeup for a cloud is a bit curious – but it may be as close to socially generated metadata as seemed reasonable for a first exploration (one that doesn’t require a substantial change in the business or publishing model).

Given the way that clouds lend themselves to showing multiple dimensions of meaning, such as change over time, I think the Times tag cloud would be more valuable if it offered the option to see all three time frames at once. I put together a quick cut and paste of a concept screen that shows this sort of layout:

Screen Concept: 3 Clouds for Different Time Frames

In an example of the rapid morphing of memes and definitions to fit shifting usage contexts (as in Thomas Vanderwal’s observations on the shifting usage of folksonomy) the NYTimes.com kept the label tag cloud, while this is more properly a weighted list: the tags shown are in fact search terms, and not labels applied to a focus of some kind by taggers.

It’s plain from the limited presence and visibility of clouds within the overall site that the staff at NYTimes.com are still exploring the value of tag clouds for their specific needs (which I think is a mature approach), otherwise I imagine the new design concept and navigation model would utilize and emphasized tag clouds to a greater degree. So far, the Times uses tag clouds only in the new “Most Popular” section, and they are offered as an alternative to the default list style presentation of popular search terms. This position within the site structure places them a few steps in, and off the standard front page-to-an-article user flow that must be one of the core scenarios supported by the site’s information architecture.

NYTimes.com User Flow to Tag Cloud

Still, I do think it’s a clear sign of increasing awareness of the potential strength of tag clouds as a way of visualizing semantic information. The Times is an established entity (occasionally serving as the definition of ‘the establishment’), and so is less likely to endanger established relationships with customers by changing its core product across any of the many channels used for delivery.

Questions of risk aside, tag clouds (here I mean any visualization of semantic metadata) couLd be a very effective way to scan the headlines for a sense of what’s happening at the moment, and the shifting importance of topics in relation to on another. With a tag cloud highlighting “immigration”, “duke”, and “judas”, visitors can immediately begin to understand what is newsworthy – at least in the minds of NYTimes.com readers.

At first glance, lowering the amount of time spent reading the news could seem like a strong business disincentive for using tag clouds to streamline navigation and user flow. With more consideration, I think it points to a new potential application of tag clouds to enhance comprehension and findability by giving busy customers powerful tools to increase the speed and quality of their judgments about what to devote their attention to in order to acheive understanding greater depth. In the case of publications like the NYTimes.com, tag clouds may be well suited for conveying snapshots or summaries of complex and deep domains that change quickly (what’s the news?), and offering rapid navigation to specific areas or topics.

A new user experience that offers a variety of tag clouds in more places might allow different kinds of movement or flow through the larger environment, enabling new behaviors and supporting differing goals than the current information architecture and user experience.

Possible Screen Flow Incorporating Clouds

Stepping back from the specifics of the design, a broader question is “Why tag clouds now?” They’re certainly timely, but that’s not a business model. This is just speculation, but I recall job postings for an Information Architect position within the NYTimes.com group on that appeared on several recruiting websites a few months ago – maybe the new team members wanted or were directed to include tag clouds in this design? If any of those involved are allowed to share insights, I’d very much like to hear the thoughts of the IAs / designers / product managers or other team members responsible for including tag clouds in the new design and structure.

And in light of Mathew Patterson’s comments here about customer acceptance of multiple clouds in other settings and contexts (priceline europe), I’m curious about any usability testing or other user research that might have been done around the new design, and any the findings related to tag clouds.

I believe the value of second generation clouds will be to offer ready navigation and access to deep, complex landscapes of meaning built up from the cumulative semantic information contained in many interconnected tag clouds. I’d like share some thoughts on this idea; I’ll split the discussion into two posts, because there’s a fair amount of material.

In a previous post on tag clouds, I suggested that the great value of first generation tag clouds is their ability to make concepts and metadata – semantic fields – broadly accessible and easy to understand and work with through visualization. I believe the shift in the balance of roles and value from first to second generation reflects natural growth in cloud usage and awareness, and builds on the two major trends of tag cloud evolution: enhanced visualization and functionality for working with clouds, and provision of extensive contextual information to accompany tag clouds.

Together, these two growth paths allow cloud consumers to follow the individual chains of understanding that intersect at connected clouds, and better achieve their goals within the information environment and outside. Fundamentally, I believe the key distinctions between first and second generation clouds will come from the way that clouds function simultaneously as visualizations and navigation mechanisms, and what they allow navigation of – landscapes of meaning that are rich in semantic content of high value.

For examples of both directions of tag cloud evolution coming together to support navigation of semantic landscapes, we can look at some of the new features del.icio.us has released in the past few months. I’ve collected three versions of the information architecture of the standard del.icio.us URL details page from the past seven months as an example of evolution happening right now.

The first version (screenshot and breakdown in Figure 1) shows the URL details page sometime before August 15th, 2005, when it appeared on Matt McAlister’s blog.

Figure 1: Del.icio.us URL Page – August 2005

The layout or information architecture is fairly simple, offering a list of the common tags for the url / focus, a summary of the posting history, and a more detailed listing of the posting history that lists the dates and taggers who bookmarked the item, as well as the tags used for bookmarking. There’s no cloud style visualization of the tags attached to this single focus available: at this time, del.icio.us offered a rendered tag cloud visualization at the aggregate level for the whole environment.

Environment and system designers know very well that as the scope and complexity of an environment increase – in this case, the number of taggers, focuses, and tags, plus their cumulative histories – it becomes more important for people to be explicitly aware of the context of any item in order to understand it properly. Explicit context becomes more important because they can rely less and less on implicit context or assumptions about context based on the universal aspects of the environment. This is how cloud consumers’ needs for clearly visible and accessible chains of understanding drives the features and capabilities of tag clouds. Later versions of this page addresses these needs in differing ways, with differing levels of success.

Figure 2 shows a more recent version of the del.licio.us history for the Ma.gnolia.com service. This screenshot taken about ten days ago in early March, while I was working on a draft of this post.

Figure 2: Del.icio.us URL Page – Early March 2006

Key changes from the first version in August to this second version include:

The most important change in this second version is the removal of the individual sets of tags from the Posting History. Separating the tags applied to the focus from associaton with the individual taggers that chose them strips them of an important layer of context. Removing the necessary context for the tag cloud breaks the chain of understanding (Figure 3) linking taggers and cloud consumers, and obscures or increases the costs of the social conceptual exchange that is the basic value of del.icio.us to its many users. In this version, cloud consumers consumers reading the URL details page can only find specific taggers based on the concepts they’ve matched with this focus by visiting or navigating to each individual taggers’ area within the larger del.icio.us environment one at a time.

Figure 3: Chain of Understanding

The switch to rendering the Common Tags block as a tag cloud is also important, as an indicator of the consistent spread of clouds to visualize semantic fields, and their growing role as navigation tools within the larger landscape.

The User Notes are a good example of an attempt to provide additional contextual information with (potentially) high value. User Notes are created by users exclusively for the purpose of providing context. The other forms of context shown in the new layout – the Posting History, Related Items – serve a contextual function, but are not created directly by users with this goal in mind. The difference between the two purposes for these items undoubtedly influences the way that people create them, and what they create: it’s a question that more detailed investigations of tagging practices will surely examine.

The third version of the same URL history page, shown in Figure 4, was released very shortly after the second, proving tag cloud evolution is happening so quickly as to be difficult to track deliberately on a broad scale.

Figure 4: Del.icio.us URL Page – March 2006 #2

This version changes the content and layout of the Posting History block, restoring the combined display of individual taggers who tagged the URL, with the tags they applied to it, in the order in which they tagged the URL for the first time.
The third version makes two marked improvements over the first and second versions:

These three different versions of the del.icio.us URL details page show that the amount and type of contextual information accompanying a single focus is increasing, and that the number of concrete navigable connections to the larger semantic landscape of which the focus is one element also increasing

Overall, it’s clear that clouds are quickly emerging as navigation tools for complex landscapes of meaning, and that cloud context has and will continue to become more important for cloud creation and use.

And so before discussing the context necesary for clouds and the role of clouds as navigation aids in more detail, it will be helpful to get an overview of landscapes of meaning, and how they arise.

Landscapes of Meaning
A landscape of meaning is a densely interconnected, highly valuable, extensive information environment rich in semantic content that is created by communities of taggers who build connected tag clouds. In the early landscapes of meaning emerging now, a connection between clouds can be a common tag, tagger, or focus: any one of the three legs of the Tagging Triangle required for a tag cloud (more on this below). Because tag clouds visualize semantic fields, connected tag clouds visualize and offer access to connected semantic fields, serving as bridges between the individual accumulations of meaning each cloud contains.

Connecting hundreds of thousands of individually created clouds and fields, as del.icio.us has enabled social bookmarkers to do by providing necessary tools and infrastructure, creates a very large information environment whose terrain or geography is composed of semantic information. Such a semantic landscape is a landscape constructed or made up of meaning. It is an information environment that allows people to share concepts or for social purposes of all kinds, while supported with visualization, contextual information, functionality, and far-ranging navigation capabilities.

The flickr Landscape
flickr is a good example of a landscape of meaning that we can understand as a semantic landscape. In a previous post on tag clouds, I considered the flickr all time most popular tags cloud (shown in Figure 5) in light of the basic structure of clouds:

“The flickr style tag cloud is …a visualization of many tag separate clouds aggregated together. …the flickr tag cloud is the visualization of the cumulative semantic field accreted around many different focuses, by many people. …the flickr tag cloud functions as a visualization of a semantic landscape built up from all associated concepts chosen from the combined perspectives of many separate taggers.”

Figure 5: The flickr All Time Most Popular Tags Cloud

From our earlier look at the structure of first generation tag clouds we know a tag cloud visualizes a semantic field made up of concepts referred to by labels which are applied as tags to a focus of some sort by taggers.
Based on our understanding of the structure of a tag cloud as having a single focus, the flickr cloud shows something different because it includes many focuses. The flickr all time most popular tags cloud combines all the individual tag clouds around all the individual photos in flickr into a single visualization, as Figure 6 shows.

Figure 6: The flickr Landscape of Meaning

This means the flickr all time most popular tags cloud is in fact a visualization of the combined semantic fields behind each of those individual clouds. It’s quite a bit bigger in scope than a traditional single focus cloud. Because the scope is so large, the amount of meaning it summarizes and conveys is tremendous. The all time most popular tags cloud is in fact a historic window on the current and historical state of the semantic landscape of flickr as a whole.

This is where context becomes critical to the proper understanding of a tag cloud. The cloud title “All time most popular tags” sets the context for this tag cloud, within the boundaries of the larger landscape environment defined and communicated by flickr’s user epxerience. Without this title, the cloud is meaningless despite the large and complex semantic landscape – all of the information environment of flickr – it visualizes so effectively, because cloud consumers cannot retrace a complete chain of understanding to correctly identify the cloud’s origin.

flickr – 1st Generation Landscape Navigation
The flickr cloud is a powerful navigation mechanism for quickly and easily moving about within the landscape of meaning built up by all those thousands and thousands of individual clouds. Still, because it is a first generation cloud, we cannot directly follow any of the many individual chains of understanding connecting this cloud’s tags back to specific taggers, or the concepts they associate with specific photos or focuses. In this visualization, the group’s understanding of meaning is more important than any individual’s understanding. And so the flickr cloud does not yet allow us comprehensive navigation of the underlying semantic landscape illustrated in Figure 6 (chains of understanding suggested in light green). The flickr cloud also remains a first generation tag cloud because users cannot control its context.

Figure 7: A Semantic Landscape

Even so, these navigational and contextual needs will help identify the way that users rely on clouds to work in landscapes of meaning.

Growth of Landscapes
Landscapes of meaning like flickr, del.icio.us, or the burgeoning number of social semantic business ventures debuting as I write – typically grow from the bottom up, emerging as dozens or thousands of individual tag clouds created for different reasons by different taggers coincidentally or deliberately interconnect and overlap, all of this happening through a variety of social mechanisms. Taggers typically create connected or overlapping tag clouds one at a time, adding tags, focuses, and taggers (by creating new accounts) in the ad hoc fashion of open networks and architectures. But first we should look at the Tagging Triangle to understand the most basic elements of a tag cloud.

The Tagging Triangle
To make a tag cloud, you have to have three elements: a focus, a tagger, and a(t least one) tag. I call this the Tagging Triangle, illustrated in Figure 8. In the most common renderings of familiar tag clouds, one or two of these elements are often implied but not shown: yet all three are always present.

This illustration shows a cloud of labels, not tags, because a rendered cloud is really a list of labels. The labels shown in most first generation clouds are often tags, but structurally they could also be a set of names for taggers, as in the del.icio.us posting history block proto-cloud we saw above, or a set of focuses as in the ‘Inverted Cloud’ I suggested.
Figure 8: The Tagging Triangle

An Example Landscape
A simple example of the growth of semantic landscapes leads naturally to the discussion of specific ways that tag clouds will enable navigation within large landscapes of meaning.

Figure 9 shows the tag cloud accreted around a single focus. This cloud includes some of the tags that Tagger 1 has used in total across all the tag clouds she’s created (those other clouds aren’t shown). We’ll assume that she’s created other clouds for other focuses.

Figure 9: A Single Tag Cloud

When a second person, Tagger 2, tags that same focus (again with a subset of the total set of all his tags), and some of those tags are the same as those used for this focus by Tagger 1, their individual tag clouds for this focus (shown by the dashed line in the cumulative tag cloud) connect via the common tags, and the cumulative cloud grows. If any of the tags from their total sets are the same, but are not used for this focus, they form another connection between the two taggers. Figure 10 shows two individual clouds connected in both these ways.

Figure 10: Two Connected Clouds

When a third tagger adds a third cloud with common tags and unique tags around the same focus, the cumulative cloud grows, and the number of both kinds of connections between tags and taggers grows. Figure 11 shows three connected clouds.

Figure 11: Connected Clouds

Every tag cloud visualizes a semantic field, and so the result of this bottom up growth is a series of interlinked semantic fields centered around a common focus, as Figure 12 shows. Since semantic fields are made of concepts, linked fields result in linked concepts.

Figure 12: Connected Semantic Fields

The total number and the variety of kinds of interconnections amongst these three taggers, their tags, and a single focus is remarkable. As this simple example shows, the total number and density of connections linking even a moderate size population of taggers, tags, and focuses could quickly become very large. This increased scale drives qualitative and quantitative topology changes in the network that permit a landscape of meaning to emerge from connected semantic fields.

Landscapes And Depth
The accumulation of connections and concepts creates a landscape of meaning with real depth; but it’s the depth of a landscape that drives its value. For this discussion, I’m defining depth loosely as the amount of semantic information or the density of the semantic field either across the whole landscape, or at a chosen point.

Value of course is a very subjective judgement. In participatory economies like that of del.icio.us, the value to individual users is predominantly one of loosely structured semantic exchange based on accumulation of collective value through shared individual efforts. From a business viewpoint, a group of investors and yahoo as a buyer saw considerable value in the emergent landscape and / or other kinds of assets

To make the idea of depth a bit clearer, Figure 13 illustrates two views of a semantic landscape built up by the overlap of tag clouds. The aerial view shows the contents, distribution, and overlap of a number of tag clouds around a set of focuses. The horizon view shows the depth of the semantic field for each focus, based on the amount of overlap or connection between the cloud around that focus and all the other clouds.

Figure 13: Semantic Landscape Depth Views

Of course this is only a conceptual way of showing the cumulative semantic information that makes up a landscape of meaning, so it does not address the relative value of this information. Plainly some indication of the quality of the semantic information in a landscape is critical important to measurements of both depth and value. Metrics for quality could come from a combination of assessment of the diversity and granularity of the tag population for the focus, benchmarks for the domain of the focus and taggers (healthcare industry), and an estimate on the maturity of the domain, the focus, and the tag clouds in the semantic landscape.
Looking ahead, it’s likely that accepted metrics for defining and describing the depth, value, and characteristics of semantic fields and landscapes will emerge as new combinations of some of the measurements used now in the realms of cognitive linguistics, set theory, system theory, topology, information theory, and quite a few other disciplines besides.

In Part Two
The second post in this series of two will follow several of the topics introduced here to conclusion, as well as cover some new topics, including:

• How chains of understanding shape needs for cloud context and navigation paths
• How the tagging triangle will define navigation within landscapes of meaning
• The emergence of stratification in landscapes of meaning
• The idea that clouds and landscapes have a shape which conveys meaning and value
• The kinds of contextual information and controls necessary for navigation and social exchanges

Watching Navigation Follow Chains of Understanding
I’ll close with a screencast put together by Jon Udell that captures a wide ranging navigation path through the del.icio.us landscape.

Lets build on the analysis of tag clouds from Tag Clouds Evolve: Understanding Tag Clouds, and look ahead at what the near future may hold for second generation tag clouds (perhaps over the next 12 to 18 months). As you read these predictions for structural and usage changes, keep two conclusions from the previous post in mind: first, adequate context is critical to sustaining the chain of understanding necessary for successful tag clouds; second, one of the most valuable aspects of tag clouds is as visualizations of semantic fields.

Based on this understanding, expect to see two broad trends second in generation tag clouds.
In the first instance, tag clouds will continue to become recognizable and comprehensible to a greater share of users as they move down the novelty curve from nouveau to known. In step with this growing awareness and familiarity, tag cloud usage will become:
1. More frequent
2. More common
3. More specialized
4. More sophisticated

In the second instance, tag cloud structures and interactions will become more complex. Expect to see:
1. More support for cloud consumers to meet their needs for context
2. Refined presentation of the semantic fields underlying clouds
3. Attached controls or features and functionality that allow cloud consumers to directly change the context, content, and presentation of clouds

Together, these broad trends mean we can expect to see a second generation of numerous and diverse tag clouds valued for content and capability over form. Second generation clouds will be easier to understand (when designed correctly…) and open to manipulation by users via increased functionality. In this way, clouds will visualize semantic fields for a greater range of situations and needs, across a greater range of specificity, in a greater diversity of information environments, for a greater number of more varied cloud consumers.

Usage Trends
To date, tag clouds have been applied to just a few kinds of focuses (links, photos, albums, blog posts are the more recognizable). In the future, expect to see specialized tag cloud implementations emerge for a tremendous variety of semantic fields and focuses: celebrities, cars, properties or homes for sale, hotels and travel destinations, products, sports teams, media of all types, political campaigns, financial markets, brands, etc.

From a business viewpoint, these tag cloud implementations will aim to advance business ventures exploring the potential value of aggregating and exposing semantic fields for a variety of strategic purposes:
1. Creating new markets
2. Understanding or changing existing markets
3. Providing value-added services
4. Establishing communities of interest / need / activity
5. Aiding oversight and regulatory imperatives for transparency and accountability.

Measurement and Insight
I think tag clouds will continue to develop as an important potential measurement and assessment vehicle for a wide variety of purposes; cloudalicious is a good example of an early use of tag clouds for insight. Other applications could include using tag clouds to present metadata in geospatial or spatiosemantic settings that combine GPS / GIS and RDF concept / knowledge structures.
Within the realm of user experience, expect to see new user research and customer insight techniques emerge that employ tag clouds as visualizations and instantiations of semantic fields. Maybe even cloud sorting?

Clouds As Navigation
Turning from the strategic to the tactical realm of experience design and information architecture, I expect tag clouds to play a growing role in the navigation of information environments as they become more common. Navigational applications comprise one of the first areas of tag cloud application. Though navigation represents a fairly narrow usage of tag clouds, in light of their considerable potential in reifying semantic fields to render them actionable, I expect navigational settings will continue to serve as a primary experimental and evolutionary venue for learning how clouds can enhance larger goals for information environments such as enhanced findability.

For new information environments, the rules for tag clouds as navigation components are largely unwritten. But many information environments already have mature navigation systems. In these settings, tag clouds will be one new type of navigation mechanism that information architects and user experience designers integrate with existing navigation mechanisms. David Fiorito’s and Richard Dalton’s presentation Creating a Consistent Enterprise Web Navigation Solution is a good framework / introduction for questions about how tag clouds might integrate into mature or existing navigation systems. Within their matrix of structural, associative and utility navigation modes that are invoked at varying levels of proximity to content, tag clouds have obvious strengths in the associative mode, at all levels of proximity to content, and potential strength in the structural mode. Figure 1 shows two tag clouds playing associative roles in a simple hypothetical navigation system.

Figure 1: Associative Clouds

I also expect navigation systems will feature multiple instances of different types of tag clouds. Navigation systems employing multiple clouds will use combinations of clouds from varying contexts (as flickr and technorati already do) or domains within a larger information environment to support a wide variety of purposes, including implicit and explicit comparison, or views of the environment at multiple levels of granularity or resolution (high level / low level). Figure 2 illustrates multiple clouds, Figure 3 shows clouds used to compare the semantic fields of a one focus chosen from a list, and Figure 4 shows a hierarchical layout of navigational tag clouds.

Figure 2: Multiple Clouds

Figure 3: Cloud Comparison Layout

Figure 4: Primary / Secondary Layout

Structural and Behavioral Trends
Let’s move on to consider structural and behavioral trends in the second generation of tag clouds.
Given the success of the simple yet flexible structure of first generation tag clouds, I expect that second generation clouds will not substantially change their basic structure. For example, tag clouds will not have to change to make use of changing tagging practices that enhance the semantic depth and quality of tags applied to a focus, such as faceted tagging, use of qualifiers, hierarchical tagging, and other forms. James Melzer identifies some best practices on del.icio.us that make considerable sense when the focus of a semantic field is a link. His recommendations include:

• Source your information with via:source_name or cite:source_name
• Create a parent categories, and thus a rudimentary hierarchy, with parent_tag/subject_tag
• Mention publications names with in:publication_name
• Flag the type of resource with .extension or =resource_type
• Use a combination of general and specific tags on every bookmark to provide both clustering and differentiation
• Use synonyms or alternate forms of tags
• Use unique or distinctive terms from documents as tags (don’t just use major subject terms)

The two element structure of first generation tag clouds can accommodate these tagging practices. However, with a semantic field of greater depth and richness available, the interactions, behaviors, and presentation of tag clouds will evolve beyond a static set of hyperlinks.

Cloud consumers’ need for better context will drive the addition of features and functionality that identify the context of a tag cloud explicitly and in detail. For example, clouds created by a defined audience will identify that audience, whether it be system administrators, freelance web designers, DJ’s, or pastry chefs rating recipes and cooking equipment and provide indication of the scope and time periods that bound the set of tags presented in the cloud. Flickr and others do this already, offering clouds of tags covering different intervals of time to account for the changing popularity of tags over their lifespan.

Moving from passive to interactive, tag clouds will allow users to change the cloud’s semantic focus or context with controls, filters, or other parameters (did someone say ‘sliders’ – or is that too 5 minutes ago…?). I’ve seen several public requests for these sorts of features, like this one: “It would be great if I could set preferences for items such as time frame or for tags that are relevant to a particular area etc or even colour the most recent tags a fiery red or remove the most recent tags.” Figure 5 shows a tag cloud with context controls attached.

Figure 5: Context Controls

Figure 6: Behavior Controls

Diversifying consumer needs and goals for way finding, orientation, information retrieval, task support, product promotion, etc., will bring about inverted tag clouds. Inverted tag clouds will center on a tag and depict all focuses carrying that tag.

Figure 7: Inverted Clouds Show Conceptually Related Focuses

In the vein of continued experiment, tag clouds will take increased advantage with RIA / AJAX and other user experience construction methods. Following this, tag clouds may take on some of the functions of known navigation elements, appearing as sub-menus / drop down menus offering secondary navigation choices.

Figure 8: Clouds As Drop Menus

Along the same lines, tag clouds will demonstrate more complex interactions, such as spawning other tag clouds that act like magnifying lenses. These overlapping tag clouds may offer: multiple levels of granularity (a general view and zoom view) of a semantic field; thesaurus style views of related concepts; parameter driven term expansion; common types of relationship (other people bought, by the same author, synonyms, previously known as, etc.)

Figure 9: Magnifying Clouds

Genres
Looking at the intersection of usage and behavior trends, I expect tag clouds will evolve, differentiate, and develop into standard genres. Genres will consist of a stable combination of tag cloud content, context, usage, functionality, and behavior within different environments. The same business and user goals that support genres in other media and modes of visualization will drive the development of these tag cloud genres. One genre I expect to see emerge shortly is the search result.

Conclusions
Reading over the list, I see this is an aggressive set of predictions. It’s fair to ask if I really have such high expectations for tag clouds? I can’t say tag clouds will take over the world, or even the Internet. But I do believe that they fill a gap in our collective visualization toolset. The quantity, quality, and relevance of semantic information in both real and virtual environments is constantly increasing. (In fact, the rate of increase is itself increasing, though that is a temporary phenomenon.) I think tag clouds offer a potential to quickly and easily support the chain of understanding that’s necessary for semantic fields across diverse kinds of focuses. There’s need for that in many quarters, and I expect that need to continue to grow.

For the moment, it seems obvious that tag clouds will spend a while in an early experimental phase, and then move into an awkward adolescent phase, as features, applications and genres stabilize in line with growing awareness and comfort with clouds in various settings.

I expect these predictions to be tested by experiments will play out quickly and in semi or fully public settings, as in the example of the dialog surrounding 83 degrees usage of a tag cloud as the sole navigation mechanism on their site that Rashmi Sinha’s post The tag-cloud replaces the basic menu – Is this a good idea? kicked off recently.

My answer to this question is that replacing all navigation menus with a tag cloud is only a good idea under very limited circumstances. It’s possible that 83 Degrees may be one of these limited instances. Startups can benefit considerably from any positive attention from the Web’s early adopter community (witness Don’t Blow Your Beta by Michael Arrington of Techcrunch).

The page’s designer said, “In this case it was done as a design/marketing effort and not at all for UI”. Since attracting attention was the specific purpose, I think the result is a success. But it’s still an experimental usage, and that’s consistent with the early stage of evolution / development of tag clouds in general.

I’m looking forward to what happens next…

Zeldman jokingly called tag clouds “the new mullets” last year. At the time, I think he was taken a bit by surprise by the rapid spread of the tag cloud (as many people were). A big year later, it looks like this version of the world’s favorite double duty haircut will stay in fashion for a while. Zeldman was discussing the first generation of tag clouds. I have some ideas on what the second generation of tag clouds may look like that will conclude this series of two essays. These two pieces combine ideas brewing since the tagging breakout began in earnest this time last year, with some predictions based on recent examples of tag clouds in practice.

Update: Part two of this essay, Second Generation Tag Clouds, is available.

This first post lays groundwork for predictions about the second generation of tag clouds by looking at what’s behind a tag cloud. I’ll look at first generation tag clouds in terms of their reliance on a “chain of understanding” that semantically links groups of people tagging and consuming tags, and thus underlies tagging and social metadata efforts in general. I’ll begin with structure of first generation tag clouds, and move quickly to the very important way that tag clouds serve as visualizations of semantic fields.

Anatomy of a Tag Cloud
Let’s begin with the familiar first generation tag cloud. Tag clouds (here we’re talking about the user experience, and not the programmatic aspects) commonly consist of two elements: a collection of linked tags shown in varying fonts and colors to indicate frequency of use or importance, and a title to indicate the context of the collection of tags. Flickr’s tags page is the iconic example of the first generation tag cloud. Screen shots of several other well known tag cloud implementations show this pattern holding steady in first generation tagging implementations such as del.icio.us and technorati, and in newer efforts such as last.fm and ma.gnolia.
Wikipedia’s entry for tag cloud is quite similar, reading, “A tag cloud (more traditionally known as a weighted list in the field of visual design) is a visual depiction of content tags used on a website. Often, more frequently used tags are depicted in a larger font or otherwise emphasized, while the displayed order is generally alphabetical… Selecting a single tag within a tag cloud will generally lead to a collection of items that are associated with that tag.”

In terms of information elements and structure, first generation tag clouds are low complexity. Figure 1 shows a schematic view of a first generation tag cloud. Figures 2 through 5 are screenshots of well-known first generation tag clouds.

Figure 1: Tag Cloud Structure

Figure 2: last.fm

Figure 3: technorati

Figure 4: del.icio.us

Figure 5: Ma.gnolia

Tag Clouds: Visualizations of Semantic Fields
The simple structure of first generation tag clouds allows them to perform a very valuable function without undue complexity. That function is to visualize semantic fields or landscapes that are themselves part of a chain of understanding linking taggers and tag consumers. This is a good moment to describe the “chain of understanding”. The “chain of understanding” is an approach I use to help identify and understand all the different kinds of people and meaning, and the transformations and steps involved in passing that meaning on, that must work and connect properly in order for something to happen, or an end state to occur. The chain of understanding is my own variation / combination of common cognitive and information flow mapping using a scenario style format. I’ve found the term resonates well with clients and other audiences outside the realm of IA.

How does the chain of understanding relate to tag clouds? The tags in tag clouds originate directly from the perspective and understanding of the people tagging, but undergo changes while becoming a tag cloud. (For related reading, see Rashmi Sinha’s A social analysis of tagging which examines some of the social mechanisms underlying the activity of tagging.) Tag clouds accrete over time when one person or a group of people associate a set of terms with a focus of some sort; a photo on flickr, a URL / link in the case of del.icio.us, an album or song for last.fm. As this list shows, a focus can be anything that can carry meaning or understanding. The terms or tags serve as carriers and references for the concepts each tagger associates with the focus. Concepts can include ideas of aboutness, origin, or purpose, descriptive labels, etc. While the concepts may change, the focus remains stable.

What’s key is that the tag is a reference and connection to the concept the tagger had in mind. This connection requires an initial understanding of the focus itself (perhaps incorrect, but still some sort of understanding), and the concepts that the tagger may or may not choose to associate with the focus. And this is the first step in the chain of understanding behind tag clouds, as shown in Figure 6.

Figure 6: Origin: Focus and Concepts

As a result, tag clouds are more than collection of descriptive or administrative terms attached to a link, or other sort of focus. The tag is a sort of label that references a concept or set of concepts. A cloud of tags is then a collection of labels referring to a cluster of aggregated concepts. The combination of tags that refer to concepts, with the original focus, creates a ‘semantic field’. A semantic field is the set of concepts connected to a focus, but in a form that is now independent of the originating taggers, and available to other people for understanding. In this sense, a semantic field serves as a form of reified understanding that the taggers themselves – as well as others outside the group that created the semantic field – can now understand, act on, etc. (This speaks to the idea that information architecture is a discipline strongly aimed at reification, but that’s a different discussion…). Figure 7 shows this second step in the chain of understanding; without it, there is no semantic field, and no tag cloud can form. And now because this post is written from the viewpoint of practical implications for tag cloud evolution, I’m going to hold the definition and discussion of a semantic field and focus, before I wander off track into semiotics, linguistics, or other territories. The most important thing to understand is that *tag clouds comprise visualizations of a semantic field*, as we’ve seen from the chain of understanding.
Figure 7: Semantic Field

I believe tag clouds are revolutionary in their ability to translate the concepts associated with nearly anything you can think of into a collectively visible and actionable information environment, an environment that carries considerable evidence of the original understandings that precede and inform it. In a practical information architecture sense, tag clouds can make metadata – one of the more difficult and abstract of the fundamental concepts of the digital universe for the proverbial person on the street – visible in an easily understood fashion. The genius of tag clouds is to make semantic concepts, the frames of understanding behind those concepts, and their manifestation as applied metadata tangible for many, many people.

Figure 8: Semantic Field As Tag Cloud

With this notion of a tag cloud as a visualization of a semantic field in mind, let’s look again at an example of a tag cloud in practice. The flickr style tag cloud (what I call a first generation tag cloud) is in fact a visualization of many tag separate clouds aggregated together. Semantically then, the flickr tag cloud is the visualization of the cumulative semantic field accreted around many different focuses, by many people. In this usage, the flickr tag cloud functions as a visualization of a semantic landscape built up from all associated concepts chosen from the combined perspectives of many separate taggers.

To summarize, creating a tag cloud requires completion of the first three steps of the chain of understanding that supports social metadata. Those steps are:
1. Understanding a focus and the concepts that could apply that focus
2. Accumulating and capturing a semantic field around the focus
3. Visualizing the semantic field as a tag cloud via transformation
The fourth step in this chain involves users’ attempts to understand the tag cloud. For this we must introduce the idea of context, which addresses the question of which original perspectives underlie the semantic field visualized in a tag cloud, and how those concepts have changed before or during presentation.

How Cloud Consumers Understand Tag Clouds
Users need to put a given tag cloud in proper context in order to understand the cloud effectively. Their end may goals may be finding related items, surveying the thinking within a knowledge domain, identifying and contacting collaborators, or some other purpose, but it’s essential for them to understand the tags in the cloud to achieve those goals. Thus whenever a user encounters a tag cloud, they ask and answer a series of questions intended to establish the cloud’s context and further their understanding. Context related questions often include “Where did these tags come from? Who applied them? Why did they choose these tags, and not others? What time span does this tag cloud cover?” Context in this case means knowing enough about the conditions and environment from which the cloud was created, and the decisions made about what tags to present and how to present them. Figure 9 summarizes the idea of context.

Figure 9: Cloud Context

Once the user or consumer places the tag cloud in context, the chain of understanding is complete, and they can being to use or work with the tag cloud. Figure 10 shows the complete chain of understanding we’ve examined.
Figure 10 Chain of Understanding

In part two, titled “Second Generation Tag Clouds”, I’ll share some thoughts on likely ways that the second generation of tag clouds will evolve in structure and usage in the near future, based on how they support a chain of understanding that semantically links taggers and tag cloud consumers. Context is the key for tag cloud consumers, and we’ll see how it affects the likely evolution of the tag cloud as a visualization tool.

Update: Part two Second Generation Tag Clouds is available

On Saturday and Sunday, I took advantage of the Blizzard of ’06 to:

1. imple­ment a tag cloud for navigation
2. tag all posts with sub­ject metadata
3. rebuild the some­what creaky col­lec­tion of stylesheets behind JL.com
4. add a recent com­ments tile

(And people say I don’t know how to have a good time…?)

The tag cloud is powered by the MoveableType plugin Tags.App. New stylesheets are loosely based on an OpenSource template from www.oswd.org called Phenom.

Between trips outside to shovel, I forgot to upload one of the new .css files. Following that, some Notes apologists justly sent me to school for displaying my comments in cripplingly small text font.

Thanks to the Notes faithful for the feedback, and condolences to any and all who contracted eye strain as a result.

Looks like Dave Sifry at Technorati has drunk the Clay Shirky Koolaid on tagging and social bookmarking. Here’s something from Dave’s posting State of the Blogosphere, August 2005, Part 3: Tags, that shows he’s clearly joined the academy of received ideas.
“Unlike rigid taxonomy schemes that many people dislike using, the ease of tagging for personal organization with social incentives leads to a rich and discoverable system, often called a folksonomy. Intelligence is provided by real people from the bottom-up to aid social discovery. And with the right tag search and navigation, folksonomy may outperform more structured approches to classification, as Clay Shirky points out…”

I’m disappointed to see this. The quality level of Shirky’s thinking and writing related to tagging is generally low; too often he’s so completely off the mark with much of what he’s said about tagging, social bookmarking, and categorization in general that his main contribution is in lending a certain amount of attention by virtue of name recognition to a subject that used to be arcane.

There’s little need to rehash the many, many individual weaknesses in Shirky’s writings, just one example of which is his establishment of a false dichotomy separating structured categorization systems and social tagging practices. Broadly, his approach and rhetoric show strong influence from anarchism, and utopian social theory.

From Shirky:
“There is no fixed set of categories or officially approved choices. You can use words, acronyms, numbers, whatever makes sense to you, without regard for anyone else’s needs, interests, or requirements.”
Further, “…with tagging, anyone is free to use the words he or she thinks are appropriate, without having to agree with anyone else about how something “should” be tagged.”

Building back on the criticique of computerization, it’s clear that Shirky uses rhetorical strategies and positions from both technological utopianism and anti-utopianism.

Here’s Professor Rob Kling on technological utopianism:
“Utopian images are common in many books and articles about computerization in society written by technologists and journalists. I am particularly interested in what can be learned, and how we can be misled, by a particular brand of utopian thought — technological utopianism. This line of analysis places the use of some specific technology, such as computers, nuclear energy, or low-energy low-impact technologies, as key enabling elements of a utopian vision. Sometimes people will casually refer to exotic technologies — like pocket computers which understand spoken language — as “utopian gadgets.”

Technological utopianism does not refer to these technologies with amazing capabilities. It refers to analyses in which the use of specific technologies plays a key role in shaping a benign social vision. In contrast, technological anti-utopianism examines how certain broad families of technology are key enablers of a harsher and more destructive social order.”

That Shirky would take speak from this standpoint is not a surprise; he’s identified as a “Decentralization Writer/Consultant” in the description of his session “Ontology is Overrated: Links, Tags, and Post-hoc Metadata” at etech, and it’s clear that he’s both technologist and a journalist, as Kilng identifies.

Regardless of Shirky’s bias, there is a bigger picture worth examining. Tagging or social bookmarking is one potential way for the community of social metadata system users to confront problems of individual and group information overload, via a collective and nominally unhierarchical approach to the emergent problem of information management across common resources (URIs).