Week 13: Information Condensation, Or, It's Only a Model!

This week’s reading: Borgmann, Part 2

“Holding On to Reality” is simultaneously the most general and the most personal treatment of information science I’ve yet read as part of the library science curriculum, and reading it is exhilarating. I’m not ready yet to address Ess’s analysis of Borgmann cited in the lecture notes, since Ess focuses primarily on Part 3 of Borgmann. Instead, I’ll return this week to the needs of users – a focal component of IA – and discuss what user needs have to do with Borgmann’s treatment of information.

Tying physical architecture to his discussion of the distinction between signs and things, Borgmann opines, “No design can specify its realization fully. To convey exactly as much information as the thing realized, a design would have to exhibit just as many features as the thing. But then it would be a duplicate of . . . the thing” (p. 113). That is, a fully realized (or fully imagined) thing must necessarily lose fidelity when it is condensed into a sign. Borgmann’s insight here is the very principle that makes indexers, abstracters, and information architects necessary. Many library users, and information consumers in general, need to know what they’re accessing before they access it. But to know exactly what a journal article says without reading the article is impossible by Borgmann’s principle. Users, then, need a general idea of what an article says – a low-fidelity version of the article. The job of an abstracter is to reduce a cataloged item (a thing) to an abstract of a more digestible length (a sign) with minimal loss of fidelity. A good abstracter makes it possible for users to make reasonable guesses about where a sign points without having to walk down the indicated road and see for themselves.

Labeling components of an information architecture is precisely analogous to abstracting media; it requires the same faculty of condensation, and has much the same end in mind in terms of how the user is served. Yet one difference between information architecture and physical architecture, which is Borgmann’s subject in Chapter 10, is that an edifice, once built, is stripped of the cultural signs used in its creation. The low-fidelity artifice of the blueprint outlives its usefulness, and the building’s users rarely need a blueprint to navigate the building. By contrast, abstracts and labels within an information system are useful precisely because they are condensed, and are indispensable for users long after the system goes live. There seems to be a fundamental disanalogy here: we can apprehend a building with our senses, and hence we navigate a building with the aid of natural signs (like a luggage carousel in an airport or a blackboard in a school) as well as cultural ones, while an information architecture is invisible to the senses and we can navigate it only through cultural signs and the guidance of the architect. Exploring a building is inherently an interactive experience; exploring an information architecture is not.

I think that this idea – the idea that an information architect must also be an information tour guide, providing signs that are naturally deficient in an online environment – is a key to overcoming user frustration with website interfaces and layouts. Since we cannot be physically present to help our users with their needs, our indexing and labeling functions are crucial to this aim. Just as John Harrison’s robust mechanical clock effectively condensed the vast grid of the world map into a longitude (p. 78), our navigation tools need to clearly help the user locate herself within the architecture; and rather like that map’s rigorous grid makes the sign revisable to match the thing, we should be ready to relabel our websites in a way that better matches the thing, or even revise the thing to match user expectations. This last possibility – the ability of the information architect to revise online reality for the convenience of the user – is probably the most exciting aspect of information architecture, and it might well be the subject of Borgmann’s Part 3, subtitled “Information As Reality.”

Week 7: Case Studies In Why We Need Information Architects

This week’s reading: Morville & Rosenfeld, Chapters 14 – 16

Our reading this week covered a number of “small” topics within information architecture. The authors’ writing was as engaging as always, but I don’t have much to add with respect to their subject treatments, so I’ll focus in this entry on the information architecture of Google Sets and Textmap.com.

I’ll begin with Google Sets. I went in expecting a very classy information architecture; Google, after all, has reams of experience designing IAs and first rose to prominence in part because of its clean, easy-to-use interface. I had mixed luck with Google Sets as a search tool – it managed to complete a list of Greek moon goddesses, but not a list of recent U.S. presidents – but the effectiveness of the search algorithms that power Google Sets is mostly outside the scope of IA. I noted, however, that Google Sets also failed to return any results given the names Sleepy, Dpoey, Bashfull, and Dock [all sic]. Basic spell-checking is part of the domain of IA, a relative of controlled vocabulary, and Google Sets ought to have been able to reconcile these misspellings.

On a related note, on those occasions when my searches returned no results, Google Sets gave some tips for more effective searching. This is good design – but I was surprised when the tips included “use the full name” and “try being consistent.” There’s no reason a search company with the resources, artificial intelligence, and processing power of Google shouldn’t be able to algorithmically guess that “Harvard” means the same thing as “Harvard University” even without a formal authority file. From such experiences with this tool’s limitations, I’d have to say that Google Sets doesn’t live up to its potential as what could be an interesting tool for finding related keywords for the purposes of tagging or building a controlled vocabulary.

From a navigation perspective, Google Sets is also faulty: to my surprise, I discovered that there is no way to revise one’s search from the results page – a mortal sin in a search engine! To end on a positive note, though, I enjoyed the metaphor of the Google Sets front page, which precedes each search field with a bullet point. This visual shorthand for a list effectively conveys what the user can do with this tool.

From Google Sets we turn to TextMap. TextMap’s IA is frankly baffling. Its “entity pages” are full of fascinating-looking metrics presented without explanation. Better labeling needs to be brought to bear on this site. At the very least, the user needs tooltips; the entity pages offer no explanation, for example, of what a “polarity rank” or “negative raw count” is. The former is defined in the site’s “Frequenty [sic] Asked Questions” – it involves whether the subject is regarded well or poorly – but there’s no indication of how TextMap makes that determination. Similarly, each entity page contains a “relational map” that links the central entity to related ideas, but there’s no way to know what prompts the relationships. The page for “cat,” for example, links the word to “Yusuf Islam.” I had to Google to figure out this relationship: the famous singer-songwriter Cat Stevens is a convert to this Islamic sect. Yet the “cat” page is clearly defined by TextMap’s terse scope note as “animal,” not “person,” so why do links pertaining to Cat Stevens appear here? The “cat” map also links to the name “Sparky,” and I still don’t know why. Each box in the relational map has a different shape – rectangle, oval, or hexagon – but no key is provided.

To draw lessons from the above, it seems that what we have in TextMap is information without architecture. No organizational skeleton puts content elements in a coherent order; no labeling scheme elucidates meaning; navigation is mostly unassisted by common conventions such as hyperlinking or a side menu; and even the search function is poor, failing to deliver the user directly to the desired page even when an exact match is found. Without architecture, the structure falls to the ground. I can’t think of a purpose that I’m confident TextMap would reliably serve.

One bright spot in TextMap is its fairly conscientious vocabulary control in the form of synonym rings. Sony’s entity page, for example, contains some forty synonyms with various permutations of capitalization and punctuation, including “SONY CORP.,” “Sony LLC,” and “Sony Electronics, Inc.” Not every permutation is covered, but the range is quite broad for a home-brewed project, and there are enough variations that a searcher could readily find the page by entering even an inexact synonym.

These two websites, then, each offer lessons in what not to do in building an information architecture. Google Sets reminds us of the importance of vocabulary control to help software make logical inferences about the user’s meaning; TextMap reminds us that data needs to be illuminated by architecture before it can properly be called information.

Week 3.5: The information architecture of lib.usf.edu

The previous post is my "official" entry for the week, but I thought I'd cross-post the following from a discussion board entry of mine in LIS 6260, which I'm taking concurrently with this course. The question concerned how libraries can help users get the most out of electronic resources. I applied this week's readings to the question:

IA exhorts us to think about how we present information. For example, on lib.usf.edu, we've made a number of good layout decisions. It's easy to find crucial information like hours and contact information, and we have a mostly well-organized set of hyperlinks in the body. But we've also made some questionable decisions. Why are links to Articles and E-Journals, which are information sources, in the same menu bar with links to ILL and Help, which are services? Why does the link labeled Books take us to the library catalog, which manifestly contains more than just books? Why do we redundantly link to the same pages under the heading Research Tools that we do in the menu bar, and why are the pages labeled differently in one place than in the other? These inconsistencies make it harder for users to build a mental model of the site. Other parts of the page seem to be designed for librarians rather than our colleagues in other fields whom we serve: What is the difference between a database and an e-journal? What is PRONTO? What is RefWorks? (For that matter, what is ILL?) Where will I go if I click on the Karst Information Portal? You won't find the answers to these questions without more clicking.

Anyway, my point is that our website's front page is not bad, but it could be better. The site doesn't do much to point a novice user in the right direction. Its flaws become transparent to veterans like ourselves, but there's a lot an experienced information architect could do to streamline and clarify it. We should *not* cop out by saying that instructors just don't give us the opportunity to teach students how to use the library. If our users can't figure out how to use our interface, the answer is not to ask our users to be more perfect, but to design our interface to be more humane.

Week 3: It looks nice, but does it work?

This week’s reading: Morville and Rosenfeld, Chapters 5 and 6

Our reading this week focused on two interconnected topics: how to conceptualize and group the organizational items of a system, and how to choose words or labels to represent them. It was an exciting pair of chapters, because the effectiveness of an information system like a website submits to empirical testing. That is, broadly speaking, it is actually possible to decide which of two possible schemes is better, in the sense of helping more of the users more of the time. In this comment I’ll focus on how we might apply empiricism to the ideas described in these chapters.

The text outlines organizational schemes appropriate for both exact and ambiguous searches. To use the language of my last entry, “fully realized” questions can be answered with straightforward schemes like alphabetical or chronological arrangement of data, but “fuzzy” questions – or items of information that fall into “fuzzy” categories – require more creativity in their organization. Most of the textbook’s examples of good interfaces give the user several access points in these ambiguous cases. Dell’s website (p. 65), for example, allows its customers to browse by topic (notebooks, desktops, support) or by audience (home, small business, government). A multiplicity of access points is likely to help some users and confuse others. Site analytics might help Dell empirically determine whether the former outnumber the latter.

A relevant metric of effectiveness might be the number of visitors who click on Dell’s topic links versus its audience links. A priori, I would expect that few visitors click “Home & Home Office” from the audience menu; these users are likely to have a more clearly defined need, and thus are more likely use the topic links. If analytics bear out this intuition, Dell should consider eliminating this hyperlink from its audience menu. Conflicting with this impulse, however, is the principle of comprehensiveness (p. 100): if we have special links for business and government audiences, shouldn’t we have a special link for home audiences? Retaining the “Home & Home Office” link might improve the menu’s consistency and thus help the user build a mental model of the Dell website, even if the link is rarely used.

The challenge, then, is to design an empirical test to settle the question of whether our little-used link contributes more than it detracts. A first approach might be to recruit a panel of diverse users, each with a genuine need. Through an automated survey, the website could prompt users to articulate their need. Half of these users could be directed to Dell’s usual site, while the other half are directed to a version of the site with the questionable link omitted. Their progress through the respective designs could be tracked and their success quantified through an exit survey. If one version of the site connects users with content with significantly higher consistency, and no intervening factors such as internal politics intervene, Dell should adopt the more successful architecture. This approach could be tested on micro aspects of design, such as whether to include a particular hyperlink, or on macro aspects, such as an entire top-down site redesign.

The Dell homepage reprinted in the book is dated 2006. I note with interest that in the intervening four years, Dell has given its site a complete revamp – consistent with the textbook authors’ emphasis on ongoing improvement. In 2006, the topical menu had pride of place on Dell’s site, while audience was relegated to a small-type menu of hyperlinks. By contrast, in 2010, the audience menu is splashed prominently across the top of the site; mousing over one of the labels (“For Home,” “For Small and Medium Business,” and so on) drops down a topic menu pertaining to the audience. This integration combines the advantages of both menus in a seamless way that is intuitive to Net-savvy audiences, though empirical testing could be useful to determine whether this two-layer sorting of content might be confusing to Internet novitiates.

Dell’s changes to its labels also merit attention. In 2006, the audience menu was headed “Solutions for:”, and its items were “Home & Home Office,” “Small Business,” “Medium & Large Business,” and “Government, Education, & Healthcare.” In 2010, the audience menu has no heading. The mouseover points are labeled “For Home,” “For Small & Medium Business,” “For Public Sector,” and “For Large Enterprise.” Three changes are interesting here. First is the change to the format of the list’s items. “Solutions for” rings of corporate jargon, which the text’s authors warn against (p. 85-86); Dell’s new formulation sounds much more natural. Second, we see that “Public Sector” has replaced the unwieldy “Government, Education, & Healthcare.” The latter choice is more descriptive, but the three indicated subcategories are heterogeneous; clicking this link is likely to lead us to a narrow, deep architecture where we’ll have to further specify that we work in education, then that we work in K-12 education, and so on. Public Sector, by contrast, denotes the same services more transparently – the label is effectively invisible to users who don’t need it – and the drop-down menu allows much of the disambiguation to take place in one click. Finally, we see that medium businesses have been reclassed with small businesses, while the term “large business” has been replaced with “large enterprise.” This could be an organizational change, but it’s more likely to be a labeling change; Dell has likely determined that its services for large businesses are disparate with the needs of medium-sized businesses, and has relabeled its categories to guide medium-sized business owners to the content most likely to be relevant to their need. The choice of the word “enterprise” in particular is clearly a labeling decision. “Enterprise” is an uncommon word whose connotation of scale may further help medium-sized business owners decide which menu category to pursue. All three of these changes have implications for the site’s overall architecture which could be user-tested by an experiment something like the one I described earlier.

My point in this entry is that IA guidelines are often useful, as when they suggest that we avoid jargon, but that site analytics and empirical testing are the ultimate tests of whether a site serves its users as envisioned. A building can be beautiful but uncomfortable, and a textbook-compliant website might still fail its users. In my first entry in this blog I discussed my view that interaction design is the parent discipline of information architecture. If so, then empiricism is the means by which we can determine whether IA is a properly dutiful child!