Information Architecture for the World Wide Web: Designing Large-Scale Web Sites

6.3. Designing Labels

Designing effective labels is perhaps the most difficult aspect of information architecture. Language is simply too ambiguous for you to ever feel confident that you've perfected a label. There are always synonyms and homonyms to worry about, and different contexts influence our understanding of what a particular term means. But even labeling conventions are questionable: you absolutely cannot assume that the label "main page" will be correctly interpreted by 100 percent of your site's users. Your labels will never be perfect, and you can only hope that your efforts make a difference, as measuring label effectiveness is an extremely difficult undertaking.

If it sounds to you like labeling is an art rather than a science, you're absolutely correct. And, as in all such cases, you can forget about finding incontrovertible rules, and hope for guidelines instead. Following are some guidelines and related issues that will help you as you delve into the mysterious art of label design.

6.3.1. General Guidelines

Remember that content, users, and context affect all aspects of an information architecture, and this is particularly true with labels. Any of the variables attached to users, content, and context can drag a label into the land of ambiguity.

Let's go back to the term "pitch." From baseball (what's thrown) to football (the field where it's played in the United Kingdom), from sales (what's sometimes made on the golf course) to sailing (the angle of the boat in the water), there are at least 15 different definitions, and it's hard to make sure that your site's users, content, and context will converge upon the same definition. This ambiguity makes it difficult to assign labels to describe content, and difficult for users to rely on their assumptions about what specific labels actually mean.

So what can we do to make sure our labels are less ambiguous and more representational? The following two guidelines may help.

6.3.1.1. Narrow scope whenever possible

If we focus our sites on a more defined audience, we reduce the number of possible perspectives on what a label means. Sticking to fewer subject domains achieves more obvious and effective representation. A narrower business context means clearer goals for the site, its architecture, and therefore its labels.

To put it another way, labeling is easier if your site's content, users, and context are kept simple and focused. Too many sites have tried to take on too much, achieving broad mediocrity rather than nailing a few choice tasks. Accordingly, labeling systems often cover too much ground to truly be effective. If you are planning any aspect of your site's scopewho will use it, what content it will contain, and how, when, and why it should be usederring toward simplicity will make your labels more effective.

If your site must be a jack of all trades, avoid using labels that address the entire site's content. (The obvious exception are the labels for site-wide navigation systems, which do cover the entire site.) But in the other areas of labeling, modularizing and simplifying content into subsites that meet the needs of specific audiences will enable you to design more modular, simpler collections of labels to address those specific areas.

This modular approach may result in separate labeling systems for different areas of your site. For example, records in your staff directory might benefit from a specialized labeling system that wouldn't make sense for other parts of the site, while your site-wide navigation system's labels wouldn't really apply to entries in the staff directory.

6.3.1.2. Develop consistent labeling systems, not labels

It's also important to remember that labels, like organization and navigation systems, are systems in their own right. Some are planned systems, some aren't. A successful system is designed with one or more characteristics that unify its members. In successful labeling systems, one characteristic is typically consistency.

Why is consistency important? Because consistency means predictability, and systems that are predictable are simply easier to learn. You see one or two labels, and then you know what to expect from the restif the system is consistent. This is especially important for first-time visitors to a site, but consistency benefits all users by making labeling easy to learn, easy to use, and therefore invisible.

Consistency is affected by many issues:

Style

Haphazard usage of punctuation and case is a common problem within labeling systems, and can be addressed, if not eliminated, by using style guides. Consider hiring a proofreader and purchasing a copy of Strunk & White.

Presentation

Similarly, consistent application of fonts, font sizes, colors, whitespace, and grouping can help visually reinforce the systematic nature of a group of labels.

Syntax

It's not uncommon to find verb-based labels (e.g., "Grooming Your Dog"), noun-based labels (e.g., "Diets for Dogs"), and question-based labels (e.g., "How Do You Paper-Train Your Dog?") all mixed together. Within a specific labeling system, consider choosing a single syntactical approach and sticking with it.

Granularity

Within a labeling system, it can be helpful to present labels that are roughly equal in their specificity. Exceptions (such as site indexes) aside, it's confusing to encounter a set of labels that cover differing levels of granularity. For example: "Chinese restaurants," "Restaurants," "Taquerias," "Fast Food Franchises," "Burger Kings."

Comprehensiveness

Users can be tripped up by noticeable gaps in a labeling system. For example, if a clothing retailer's site lists "pants," "ties," and "shoes," while somehow omitting "shirts," we may feel like something's wrong. Do they really not carry shirts? Or did they make a mistake? Aside from improving consistency, a comprehensive scope also helps users do a better job of quickly scanning and inferring the content a site will provide.

Audience

Mixing terms like "lymphoma" and "tummy ache" in a single labeling system can also throw off users, even if only temporarily. Consider the languages of your site's major audiences. If each audience uses a very different terminology, you may have to develop a separate labeling system for each audience, even if these systems are describing exactly the same content.

There are other potential roadblocks to consistency. None is particularly difficult to address, but you can certainly save a lot of labor and heartache if you consider these issues before you dive into creating labeling systems.

6.3.2. Sources of Labeling Systems

Now that you're ready to design labeling systems, where do you start? Believe it or not, this is the easy part. Unless you're dealing with ideas, concepts, and topics that until now were unknown to humanity, you'll probably have something to start with. And already having a few labels generally beats starting from scratch, which can be prohibitively expensive, especially with large vocabularies.

Existing labeling systems might include the labels currently on your site, or comparable or competitors' sites. Ask yourself who might have taken this on before. Study, learn, and "borrow" from what you find on other sites. And keep in mind that a major benefit of examining existing labeling systems is that they're systemsthey're more than odd, miscellaneous labels that don't necessarily fit together.

As you look for existing labeling systems, consider what works and what doesn't. Which systems can you learn from, and, perhaps more importantly, which of those labels can you keep? There are a variety of sources for labels that you should examine.

6.3.2.1. Your site

Your web site probably already has labeling systems by default. At least some reasonable decisions had to have been made during the course of the site's creation, so you probably won't want to throw all those labels out completely. Instead, use them as a starting point for developing a complete labeling system, taking into consideration the decisions made while creating the original system.

A useful approach is to capture the existing labels in a single document. To do so, walk through the entire site, either manually or automatically, and gather the labels. You might consider assembling them in a simple table containing a list or outline of each label and the documents it represents. Creating a labeling table is often a natural extension of the content inventory process. It's a valuable exercise, though we don't recommend it for indexing term vocabularies, which are simply too large to table-ize unless you focus on small, focused segments of those vocabularies.

Following is a table for the navigation system labels on jetBlue's main page.

Arranging labels in a table provides a more condensed, complete, and accurate view of a site's navigation labels as a system. Inconsistencies are easier to catch; in jetBlue's case, we encounter three variants of the company's name alone: "jetBlue," "JetBlue," and "JetBlue Airways." We find inconsistencies for a single page's labels: the contact page is labeled "Contact us," "Contacts," and "Learn more - JetBlue." Many pages don't have main headings. We encounter various other style inconsistencies that may confuse users. We may decide that, personally, we just don't like certain labels. We may also decide that some of the problems aren't worth changing. In any case, we now have a sense of the site's current labeling system and how it could be improved.

6.3.2.2. Comparable and competitive sites

If you don't have a site in place or are looking for new ideas, look elsewhere for labeling systems. The open nature of the Web allows us to learn from one another and encourages an atmosphere of benevolent plagiarism. So, just as you might view the source of a wonderfully designed page, you can "borrow" from another site's great labeling system.

Determine beforehand what your audiences' needs are most likely to be, and then surf your competitors' sites, borrowing what works and noting what doesn't (you might consider creating a label table for this specific purpose). If you don't have competitors, visit comparable sites or sites that seem to be best in class.

If you surf multiple competitive or comparative sites, you may find that labeling patterns emerge. These patterns may not yet be industry standards, but they at least can inform your choice of labels. For example, in a recent competitive analysis of eight financial services sites, "personal finance" was found to be more or less a de facto label compared to its synonyms. Such data may discourage you from using a different label.

Figure 6-12 shows labeling systems from Compaq, Gateway, Dell, and IBM, all competing in the PC business. Do you notice a trend here?

6.3.2.3. Controlled vocabularies and thesauri

Another great source is existing controlled vocabularies and thesauri (a topic we'll cover in depth in Chapter 9). These especially useful resources are created by professionals with library or subject-specific backgrounds, who have already done much of the work of ensuring accurate representation and consistency. These vocabularies are often publicly available and have been designed for broad usage. You'll find these to be most useful for populating labeling systems used for indexing content.

But here's a piece of advice: seek out narrowly focused vocabularies that help specific audiences to access specific types of content. For example, if your site's users are computer scientists, a computer science thesaurus "thinks" and represents concepts in a way your users are likely to understand, more so than a general scheme like the Library of Congress subject headings would.

A good example of a specific controlled vocabulary is the Educational Resources Information Center (ERIC) Thesaurus. This thesaurus was designed, as you'd guess, to describe the domain of education. An entry in the ERIC Thesaurus for "scholarships" is shown in Figure 6-13.

If your site has to do with education or if your audience is comprised of educators, you might start with ERIC as the source for your site's labels. You can use a thesaurus like ERIC to help you with specific labeling challenges, like determining a better variant for a particularly knotty label. You might go as far as to license the entire vocabulary and use it as your site's labeling system.

Unfortunately, there aren't controlled vocabularies and thesauri for every domain. Sometimes you may find a matching vocabulary that emphasizes the needs of a different audience. Still, it's always worth seeing if a potentially useful controlled vocabulary or thesaurus exists before creating labeling systems from scratch. Try these four excellent lists as you hunt for sources of labels:

• Taxonomy Warehouse: http://taxonomywarehouse.com/

• ThesauriOnline (American Society of Indexers): http://www.asindexing.org/site/thesonet.shtml

• Controlled vocabularies (Michael Middleton): http://sky.fit.qut.edu.au/~middletm/cont_voc.html

• Web Thesaurus Compendium (Barbara Lutes): http://www.ipsi.fraunhofer.de/~lutes/thesoecd.html

6.3.3. Creating New Labeling Systems

When there are no existing labeling systems or when you need to do more customizing than you'd expected, you face the tougher challenge of creating labeling systems from scratch. Your most important sources are your content and your site's users.

6.3.3.1. Content analysis

Labels can come directly from your site's content. You might read a representative sample of your site's content and jot down a few descriptive keywords for each document along the way. It's a slow and painful process, and it obviously won't work with a huge set of documents. If you go this route, look for ways to speed up the process by focusing on any existing content representations like titles, summaries, and abstracts. Analyzing content for candidate labels is certainly another area where art dominates science.

There are software tools now available that can perform auto-extraction of meaningful terms from content. These tools can save you quite a bit of time if you face a huge body of content; like many software-based solutions, auto-extraction tools may get you 80 percent of the way to the finish line. You'll be able to take the terms that are output by the software and use them as candidates for a controlled vocabulary, but you'll still need to do a bit of manual labor to make sure the output actually makes sense. (And it's worth noting that auto-extraction toolsand the training and tuning to make them work wellcan be quite expensive.) We provide pointers to some auto-extraction tools in Chapter 16.

6.3.3.2. Content authors

Another manual approach is to ask content authors to suggest labels for their own content. This might be useful if you have access to authors; for example, you could talk to your company's researchers who create technical reports and white papers, or to the PR people who write press releases.

However, even when authors select terms from a controlled vocabulary to label their content, they don't necessarily do it with the realization that their document is only one of many in a broader collection. So they might not use a sufficiently specific label. And few authors happen to be professional indexers.

So take their labels with a grain of salt, and don't rely upon them for accuracy. As with other sources, labels from authors should be considered useful candidates for labels, not final versions.

6.3.3.3. User advocates and subject matter experts

Another approach is to find advanced users or user advocates who can speak on the users' behalf. Such people may include librarians, switchboard operators, or subject matter experts (SMEs) who are familiar with the users' information needs in a larger context. Some of these peoplereference librarians, for examplekeep logs of what users want; all will have a good innate sense of users' needs by dint of constant interaction.

We found that talking to user advocates was quite helpful when working with a major healthcare system. Working with their library's staff and SMEs, we set out to create two labeling systems, one with medical terms to help medical professionals browse the services offered by the healthcare system, the other for the lay audience to access the same content. It wasn't difficult to come up with the medical terms because there are many thesauri and controlled vocabularies geared toward labeling medical content. It was much more difficult to come up with a scheme for the layperson's list of terms. There didn't seem to be an ideal controlled vocabulary, and we couldn't draw labels from the site's content because it hadn't been created yet. So we were truly starting from scratch.

We solved this dilemma by using a top-down approach: we worked with the librarians to determine what they thought users wanted out of the site. We considered their general needs, and came up with a few major ones:

1. They need information about a problem, illness, or condition.

2. The problem is with a particular organ or part of the body.

3. They want to know about the diagnostics or tests that the healthcare professionals will perform to learn more about the problem.

4. They need information on the treatment, drug, or solution that will be provided by the healthcare system.

5. They want to know how they can pay for the service.

6. They want to know how they can maintain their health.

We then came up with basic terms to cover the majority of these six categories, taking care to use terms appropriate to this audience of laypersons. Here are some examples:

By starting with a few groupings, we were able to generate labels to support indexing the site. We knew a bit about the audience (laypersons), and so were able to generate the right kinds of terms to support their needs (e.g., leg instead of femur). The secret was working with people (in this case, staff librarians) who were knowledgeable about the kind of information the users want.

6.3.3.4. Directly from users

The users of a site may be telling you, directly or indirectly, what the labels should be. This isn't the easiest information to get your hands on, but if you can, it's the best source of labeling there is.

6.3.3.4.1. Card sorting

Card sort exercises are one of the best ways to learn how your users would use information. (Card sorting methodologies^[] are covered more extensively in Chapter 10.) There are two basic varieties of card sorts: open and closed. Closed card sorts provide subjects with existing categories and ask them to sort content into those categories. At the start of a closed card sort, you can ask users to explain what they think each category label represents and compare these definitions to your own. Both approaches are useful ways to determine labels, although they're more appropriate for smaller sets of labels such as those used for navigation systems.

^[] We also anticipate that Donna Maurers book, Card Sorting: The Book will be quite helpful here; it will be published by Rosenfeld Media in early 2007 (http://www.rosenfeldmedia.com/books/cardsorting).

In the example below, we asked subjects to categorize cards from the owner's section of a site for a large automotive company (let's call it "Tucker"). After we combined the data from this open card sort, we found that subjects labeled the combined categories in different ways. "Maintenance," "maintain," and "owner's" were often used in labels for the first cluster, indicating that these were good candidates for labels (see Table 6-1).

But in other cases, no strong patterns emerged (see Table 6-2).

In a corresponding closed card sort, we asked subjects to describe each category label before they grouped content under each category. In effect, we were asking subjects to define each of these labels, and we compared their answers to see if they were similar or not. The more similar the answers, the stronger the label.

Some labels, such as "Service & Maintenance," were commonly understood, and were in line with the content that you'd actually find listed under this category (see Table 6-3).

Other category labels were more problematic. Some subjects understood "Tucker Features & Events" in the way that was intended, representing announcements about automobile shows, discounts, and so on. Others interpreted this label to mean a vehicle's actual features, such as whether or not it had a CD player (see Table 6-4).

Card sort exercises are very informative, but it's important to recognize that they don't present labels in the context of an actual site. Without this natural context, the labels' ability to represent meaning is diminished. So, as with all other techniques, card sorts have value but shouldn't be seen as the only method of investigating label quality.

6.3.3.4.2. Free-listing

While card sorting isn't necessarily an expensive and time-consuming method, free-listing is an even lower-cost way to get users to suggest labels.^[§] Free-listing is quite simple: select an item and have subjects brainstorm terms to describe it. You can do this in person (capturing data with pencil and paper will be fine) or remotely, using a free or low-cost online-survey tool like SurveyMonkey or Zoomerang. That's really all there is to it.

^[§] The best summary of this method is Rashmi Sinha's short but highly useful article in the Februrary 2003 Boxes & Arrows, "Beyond cardsorting: Free-listing methods to explore user categorizations" (http://www.boxesandarrows.com/view/beyond_cardsorting_free_listing_methods_to_explore_user_categorizations).

Well, not quite: you'll want to consider your subjects: who (ideally representative of your overall audience) and how many (three to five may not yield scientifically significant results, but it is certainly better than nothing and may yield some interesting results). You might also consider asking subjects to rank the terms they've suggested as a way to determine which are the most appropriate.

You'll also need to choose which items to brainstorm terms for. Obviously you can only do this with a subset of your content. You could choose some representative content, such as a handful of your company's products. But even then, it'll be trickydo you choose the most popular products or the more esoteric ones? It's important to get the labeling right for your big sellers, but conventions for their labels are already fairly established. The esoteric items? Well, they're more challenging, but fewer people care about them. So you may end up with a balance among the few items you select for a free-listing exercise. This is one of those cases where the art of information architecture is at least as important as the science.

What do you do with the results? Look for patterns and frequency of usage; for example, most of your subjects use the term "cell phone" while surprisingly few prefer "mobile phone." Patterns like these provide you with a sense of how to label an individual item, but may also demonstrate the tone of users' language overall. You might note that they use jargon quite a bit, or the reverse; perhaps you find a surprising amount of acronyms in their labels, or some other pattern emerges from free-listing. The result won't be a full-fledged labeling system, but it will give you a better sense of what tone and style you should take when developing a labeling system.

6.3.3.5. Indirectly from users

Most organizationsespecially those whose sites include search enginesare sitting on top of reams of user data that describe their needs. Analyzing those search queries can be a hugely valuable way to tune labeling systems, not to mention diagnose a variety of other problems with your site. Additionally, the recent advent of folksonomic tagging has also created a valuable, if indirect, source of data on users' needs that can help information architects develop labeling systems.

6.3.3.5.1. Search-log analysis

Search-log analysis (also known as search analytics) is one of the least intrusive sources of data on the labels your site's audiences actually use. Analyzing search queries^[||] is a great way to understand the types of labels your site's visitors typically use (see Table 6-5). After all, these are the labels that users use to describe their own information needs in their own language. You may notice the use (or lack thereof) of acronyms, product names, and other jargon, which could impact your own willingness to use jargony labels. You might notice that users' queries use single or multiple terms, which could affect your own choice of short or long labels. And you might find that users simply aren't using the terms you thought they would for certain concepts. You may decide to change your labels accordingly, or use a thesaurus-style lookup to connect a user-supplied term (e.g., "pooch") to the preferred term (e.g., "dog").

^[||] Naturally, we have one more book to recommend that's not yet quite available at press time, but that should be useful nonetheless: Search Analytics for Your Site: Conversations with Your Customers, by Louis Rosenfeld and Rich Wiggins. It will be published by Rosenfeld Media and should be available in early 2007 (http://www.rosenfeldmedia.com/books/searchanalytics).

6.3.3.5.2. Tag analysis

The recent explosion in sites that employ folksonomic tagging (i.e., tags supplied by end users) means another useful indirect source of labels for you to learn from. In many of these sites, users' tags are publicly viewable. When you display them in aggregate, you'll find a collection of candidate labels that approximates the results of a free-listing exercise. Additionally, the data that comes from tag analysis can be used in much the same way as search-log analysis. Look for common terms, but also look for jargon, acronyms, and tone; even misspellings are useful if you're building a controlled vocabulary.

In the examples shown in Figures 6-14 and 6-15, you might be wondering how to develop labels for a new web-based iPod accessories store. To start, you might look at a popular folksonomic system like del.icio.us and see whether users have tagged a few common iPod accessories, and what terms they used. Let's try a pair of iPod accessories, a radio remote and a leather case. After searching both terms in del.icio.us, we found a variety of results, and chose those that had been bookmarked the most times.

Some of the tags are too broad to be particularly useful (e.g., "iPod" or "shopping"). But some will help you determine labels for categories; in the first example, "hardware" is more common than "media." Knowing that will clarify your category labeling. In the second example, you might choose "case" over the less popular "cases" as a product label.

6.3.4. Tuning and Tweaking

Your list of labels might be raw, coming straight from the content in your site, another site, your site's users, or your own ideas of what should work best. Or, it may come straight from a polished controlled vocabulary. In any case, it'll need some work to become an effective labeling system.

First, sort the list of terms alphabetically. If it's a long list (e.g., from a search log), you'll likely encounter some duplicates; remove these.

Then review the list for consistency of usage, punctuation, letter case, and so forth, considering some of the consistency issues discussed earlier in this chapter. For example, you'll remember that the label table drawn from the jetBlue web site had inconsistencies that were immediately obvious: sometimes there were periods after labels, sometimes there weren't; the usage of link labels versus the heading labels on the corresponding pages was inconsistent; and so on. This is a good time to resolve these inconsistencies and to establish conventions for punctuation and style.

Decisions about which terms to include in a labeling system need to be made in the context of how broad and how large a system is required. First, determine if the labeling system has obvious gaps. Does it encompass all the possibilities that your site may eventually need to include?

If, for example, your e-commerce site currently allows users to search only a portion of your product database, ask yourself if eventually it might provide access to all products. If you're not certain, assume it will, and devise appropriate labels for the additional products.

If the site's labeling system is topical, try to anticipate the topics not yet covered by the site. You might be surprised to see that the addition of these "phantom" labels has a large impact on your labeling system, perhaps even requiring you to change its conventions. If you fail to perform this predictive exercise, you might learn the hard way that future content doesn't fit into your site because you're not sure how to label it, or it ends up in cop-out categories such as "Miscellaneous," "Other Info," and the classic "Stuff." Plan ahead so that labels you might add in the future don't throw off the current labeling system.

Of course, this planning should be balanced with an understanding of what your labeling system is there to accomplish today. If you try to create a labeling system that encompasses the whole of human knowledge (instead of the current and anticipated content of your web site), don't plan on doing anything else for the rest of your life. Keep your scope narrow and focused enough so that it can clearly address the requirements of your site's unique content, the special needs of its audiences, and the business objective at hand, but be comprehensive within that well-defined scope. This is a difficult pursuit, to be sure; all balancing acts are. Consider it justification #64 for information architectslike yourselfto be paid well.

Finally, remember that the labeling system you launch will need to be tweaked and improved shortly thereafter. That's because labels represent a relationship between two thingsusers and contentthat is constantly morphing. Stuck between two moving targets, your labeling system will also have to change. So be prepared to perform user tests, analyze search logs on a regular basis, and adjust your labeling system as necessary.