Faceted Browsing and Taxonomy

This entry is a work in progress intended as a tentative study of current use of faceted (guided) navigation in e-commerce settings and how it exposes underlying taxonomy to the user. This blog entry is NOT a critique of the sites discussed here but rather an exploration of navigation paths, taxonomy facilitated browsing and assumptions made regarding the inherent use of underlying information architecture and its impact on clarity and usability (directly impacting conversion and retention rates).

1. Lowe's.com [Captured January 2009]

There are several ways to navigate the site by browsing. The left column provides groupings that parallel the top horizontal menu. In the left column the user see items grouped by Departments and below that, items grouped by Rooms. Departments maps to the store or correspond to a mental model a user might have of the store, and Rooms map to a home or correspond to a mental model the user might have of a home. Providing multiple browsing models is a nice feature because it supports self identification - the user benefits from a flexibility to be at their comfort level, not the site's.

One can assume that the user is more familiar with the concept of a home, so it is a pity that the navigation the user is more comfortable with is secondary to the one the store's model. On the other hand, some users may also be very familiar with the store model. For example - store clerks or customer service reps. (But in my experience the in-store terminals are generally not similar to a company's public e-commerce store). In any case, the site exposes and organizes the highest level of its taxonomy and access to its products in two ways, which increases the flexibility and and probability the user will select one path to work with and not abandon the site.

The number of items under the Rooms model is obviously much shorter than those under Departments and more over, the Laundry Room is one of the items listed right there on this first level. But, browsing top down path 1 (See image below) is the first the user will encounter, clicking the Appliance link under 'Departments'.

>>> Assumption: The user would know/guess that a dryer is an appliance.

Path 1:

L1.1 - Departments

L1.2 - Appliances (click to get to L2)

If the user is more inquisitive and visually scrolls down to the Rooms section, the obvious, explicit selection is right there. (See image

Path 2:

l1.3 - Rooms

l1.4 - Laundry Room (click to get to L2)

Both browse path involve 2 clicks, so no efficiency is gained in terms of physical effort.

Talking Taxonomy To Kids

Everyone knows that you need to use simple words when talking to little kids, so 'big' words like classification, clustering, facets and hierarchy (a distasteful term which even some grownups find difficult to spell) are out. So let's start with a tree. Imagine a tree. What's a tree really? You could say that a tree is like a 'parent', and it has 'children': A trunk that splits into several big branches that in turn split into smaller twigs that split into even smaller twigs where leaves sprout and fruit grow. If the child is really curious, you can talk about the parts of the tree that only moles could see if moles could see: The taproot which is the main root that grows vertically into the ground, the lateral roots that parallels the branches, the radicles which is just a word for small roots that parallel twigs, and the root hair zone which is like the leaves. But lets not make things complicated, because we are talking to a child who happens to speak English. We'd have to use words like stamm (trunk), zweig (branch) and zweig again for twig because it seems that the Germans don't have a special word for it, or at least that's what you get from online translators.

So. is taxonomy like a tree? wait, wait... because there is another way to describe a tree: The bole is the part of the tree between the ground and the first branch, the crown which is the part of the tree from the first branch to the top, and the top which is the highest part of the tree.

And...a tree is a plant and there are all kinds of trees and here are just a few: Redwood, Ash, Fir, Spruce, Sequoia. There are banana trees, apple trees, orange trees and how exactly is a Sequoia related to and Avocado tree? And even more: A collection of trees can form a forest, grove, garden, or park which by themselves are not just collections of trees but wider concepts. So of course, the development of a taxonomy involves research, and among other things, one can find that for many domains, especially in life sciences, law, government and many others, it is possible to start the process with an existing taxonomy, see for example the taxonomywarehouse.

It is clear that the scope of concepts in the world is endless due to the Human instinct to stereotype and classify, first explored by Aristotle, or at least - this is our first written record of an arrangement to classes, subclasses and so on, as means to understand our world. But if so, lets keep in mind that children must have an inherent understanding of classification, and by extenuation, of taxonomy. It is only a matter of vocabulary then.

Taxonomy is a communication device, a tricky one because it is important to make sure that the person that communicates the taxonomy and the audience for the taxonomy understand each other. So the first thing is to understand - who will be using your taxonomy and how.

When you talk to a child, you want to talk about the tree using words like brunch, twigs, leaves etc. and you don't want to discuss apical dominance, foliage and phloem because this is not the vocabulary of an eight years old. And since we clearly can apply elementary school education to user experience design, I would add that developing a taxonomy is art as much as it is science, and for that you can read more in Bowker and Star's great book 'Sorting Things Out'.

As a communication device, taxonomy's principal use is in navigation systems and facilitating good search results. But because a taxonomy maps to a known mental model shared by the user and the system, it is important that the appropriate taxonomy will be exposed to the user in navigation systems, drop-lists, and other actionable interface objects. Such a system allows the user to self-identify: I'm a kid, I'm a teacher or I'm a parent/guardian, and the system renders relevant taxonomies based on appropriate synonym mapping. The multidimensionality of relations within a taxonomic plane is supported by explicit content tagging as well as folksonomy - a taxonomy set by users, to provides the necessary flexibility.

Towards A Unified UI Testing Model

This entry was inspired by a post by Avinash Kaushik 'Experiment or Die. Five Reasons And Awesome Testing Ideas'.

Although 'User' is the operative word in 'User Interface', it took several decades to get usability off the ground as a service companies are willing to pay for. It is true that some companies pioneered user centered design years ago, but I think it is safe to say that the 'main street' of companies involved in any substantial software project considered (and many still do) the user interface mere eye candy. But the evidence for an evolution is the accepted legitimacy of roles such information and user experience architects, usability engineers, interface designers and so on.

As a result of the higher awareness to the UI throughout a software's life cycle, testing the UI during development is now increasingly common as the tools needed to conduct reasonable testing are more affordable, and testing goals are more practical. Consumer facing interface re/design projects are increasingly adding usability testing as part of the pre-launch process and there is certainly a shift from pseudo-scientific testing of eye-movement tracking or user response time to on-screen events, to measures of task flow efficiency and task completion success.

Usability testing software such as Morae and UserVue substantially reduce the expense and limitations of UI testing that were common just a few years ago when usability labs had to be be rented by the hour, and were extremely expensive. In early 2006 I was handed sixteen audio-cassettes of ninety minutes each after finishing a couple of days in a usability lab. The client spent over $10K for the testing and yet the budget did not allow for video taping and there was no time nor budget allocation to go over the audio tapes after the sessions. While we learned a lot form the sessions, and $10K were a drop in the bucket for a multi-million dollar project, the singularity of such an exercise turned it into an expensive line item that was difficult to sell to many clients, whose budget for UI work was limited to begin with.

The truth is that the technology was just not there in terms of computing powers for real time audio and video capture possible now, and best practices were thin, since performing lab tests was a rare occasion for most practitioners. But the big drawback, in my mind, involved limited demographic and geographic distribution of the test participant due to their need to be in a relative close proximity to the testing facility. Today, with web based testing, we re no longer limited to a physical location and are able to sample a spread that is an accurate reflection of an application's user audience. Methodologies and best practices for UI testing are evolving rapidly, and acceptance of this effort is so high such that it is no longer questioned, as long as the cost is reasonable. UI testing prior to (and maybe during) development makes all the sense.

What I often find is a reality in which organizations contract UI design services - especially interaction design and information architects. As a result, navigation systems, page layouts and behavior patterns of landing pages are set during the concept development phase. Companies will pay for some iterations of user validations, but there is always a real budget pressure to release asap and cut costs. I have yet to see a project plan that seriously accounts for sufficient exploration and testing, and have to fight for it time and again. It is not that clients don’t see the value, but they don’t want to pay unless the concept is seriously off target.

To be realistic and practical - It takes significant time and labor (=$$$) to determine and preserve patterns of consistent interaction and visual design approach and the variations possible. The efforts can be significantly bigger when you are dealing with a multi-national presence where one needs to account for many stakeholders as well as contrasting cultural sensibilities. It is very rare to have such luxury and moreover, critics may argue that the best evolution of the redesigned UI will take place in deployment, not in the 'lab'.

And so, in many cases, the UI design consulting firm leaves around deployment time after handoff to the internal development team and this is where the brand new UI begins to fall apart - there is no one internally with the skill-set, time, budget to take charge of testing the evolving interface as it is being readied for deployment. I doubt that The style guides and UI specs are used much; the cynical phrase 'No one reads' is no far off reality partially because specs are difficult to produce and hard to consume. But that is another story.

As it turns out the UI often gets tested again once in production. This is especially true for commercial B2B and B2C RIAs. However, this round of testing and decisions about modifications to the UI are often done outside of the context of usability, often, without the involvement of the UI team that architected it (due to the fact that often, the consultants who were hired to develop the applications UI are not retained after the launch. In fact - the people who do this round of testing often know very little about UI practice OR even look at the UI they test and attempt to improve.

Usability testing:

• The testing is performed by usability professionals, part of a concentrated, focused UI effort.
• The testing is typically qualitative because the sample of participants is relatively small.
• The testing is typically done on low fidelity clickable prototype, a semi-functional POC, or for redesign purposes on the deployed software.
• The testing validates the design concept and triggers stakeholders' sign-off, or guides improvements to the existing or redesigned software.

Web Analytics Conversion testing:

• The testing is performed by web analytics professionals and the effort is typically not related to a UI effort: The testing is not really focused on the user interface from a usability perspective, but from an optimization perspective.
• The testing is quantitative, based on actual web analytics data derived from deployment usage.
• The tested user interface is the production UI.

Analytics testing takes time - time to plan the testing strategy, prepare it, but most of all, time to execute and wait to see if trends are changing. We can not assume that the change will take place overnight. Is there a way to attribute time factor to the success or failure of a tested approach? Was it a single element that has contributed to the change, or is it the combination? or is it the latent impact of the brand, of market drivers, reduction of costs and so on.

During development, usability testing is iterative, fast and qualitative. Often this is where testing ends for many organizations, they stop using the consultants and move to analytics testing that is performed by web analytics consultant, or, it is likely that they will have someone in house. Analytics testing is on-going, quantitative, and can be like stabbing in the dark - trying to figure Why without tying it to usability.

Clearly a gap in the interaction design discourse when it comes to web analytics (and testing for optimization). Analytics is regarded as a ‘post’ event, not as something you can be proactive about during the design process. What I hope to see is more dialog between the user experience community and the web analytics community around practical ways to integrate testing and develop a full life cycle approach that combines usability and analytic s considerations throughout. More to come.

Tutorial: Simulate Type-Ahead Using Axure

I created this audio-visual tutorial a few months ago. Several people have asked me since to post it in this blog. Here is the link: http://www.artandtech.com/type-ahead.html

Real-Time, Quantitative Capture of User Response to Streaming Content

1. Introduction

Usability studies utilize both qualitative and quantitative methods for capturing user response to the user interface that is being tested. We can measure mouse-clicks, time on task, task completion rates and other valuable data. We can also collect verbal feedback related to ease of use, visual design, layout and other subjective responses. The processing of collected verbal data is expensive because recordings have to be transcribed, tagged and often edited for readability. This is a labor intensive process and if the testing is done with users who talk different languages, translation is also required. Moreover, even when interviews are carefully scripted and prompts are consistent, response are often difficult to reconcile:  Participant's answers can be inconsistent, vague, and generally difficult to analyze and interpret.

Verbal feedback is also used to capture participants' response to streaming content and to gage level of engagement with that content. Typically the tester pauses the media and prompts the participant for her or his opinion. The benefit of this method is that the feedback is contextually related to the content which had just been displayed and is fresh in the mind of the respondent. The disadvantage is the labor intensive post session processing and interpretation of the information gathered. Alternatively, a user can be given a questioner at the end of the streaming content. The benefit of a questioner is that it is easier to process and measure the responses, but the drawback is that the participant is not likely to recall in deep detail their response to content or their sense of engagement with content that was displayed minutes ago.

This paper describes a method I developed to capture in real-time participants' response to streaming content as well as their engagement levels throughout the presentation. The key benefits of this method are:

1. Capture in real-time users' response to streaming content such as web seminars, tutorials and demos, where the user interface itself plays a smaller role in the interaction
2. Significantly lower the time labor costs associated with processing the feedback, which may help budgeting for larger samples.
   
3. Capture response to streaming content by setting your own test pages or from any website or application.

The method involves the use of TechSmith's Morae*, which is currently the only commercial, out-of-the-box software for usability test. The method leverages Morae's capability to captures, among other things, mouse movement and mouse clicks.

2. Methods

2.1 Create your own test page/s

The first approach is to create your own test pages. This scenario works well when you:

1. Wish to hide the tested content from the associated company's identity by isolating it from the rest of the company's site and the site's URL. 
2. When you are testing several draft variations of the content, don't want to bother site admins with helping you post the stuff and need to run it locally off your machine. 

An added benefit is that you can perform the test without worrying about the quality of bandwidth in the test location, or an internet connection all together.

Some technical skills involving the creation of a standard web page are required for setting up your own test pages, but a typical page is really simple, composed of the embedded streaming content - typically a Flash file (So you will need the SWF file), and a single graphics that is used to capture the feedback for content and engagement. See image 1 below:

You need to create an image that will be used to capture the user's responses to content and the user's engagement level. This graphic can be as fancy as you wish, but my suggestion is to keep it simple and remember that the main event on the page is the streaming content, not these graphics. Here is an image I typically use:

The image is divided into 2 sections:

1. Left side - Response to content. A rating scale from 1 to 7, with 1 being "I don't care -- trivial content" to 7 being "Really important -- Tell me more!"
2. Right side - Engagement level. A rating scale from 1 to 7, with 1 being "I'm bored" to 7 being "I'm fully engaged"

Morae Study Configuration

To maximize efficiency of logging sessions in Morae Manager, it is best to prepare the study configuration in advance. See image below:

For a 7 based rating scale, prepare 7 markers for content and 7 markers for engagement and label them Content 1, Content 2, etc. Change the letter association for the markers to a sequence that will make it easy for you to use shortcut during the logging. Finally, assign a color to all content markers, and a different one to all engagement markers. This different colors will provide a clear differentiation once you finish placing all the markers.

How it Works:

Ask the user to click the relevant ratings on the content and engagement bars as the content streams. Ask the user to click as many times as makes sense. Morae captures mouse clicks on the bars, which are easy to see and log. (The red triangle in the image below is generated by Moare Recorder during the session.)

In logging the session it is possible to identify with a high degree of accuracy which section of the streaming content the participant rated, and of course, the assigned value. With a big enough sample rate you can get a good insight into participants opinion about the content -- both narration and visuals, as well as their engagement level throughout the streaming.

Some production tips:

1. For the screen to be aesthetically pleasing and professionally looking, I adjust the width of the image so that it is the same as the width of the embedded content I'm testing.
2. The buttons on the bar should be clear and easy to see, and easy to click on. 
3. The labels should be clear and easy to read
4. This is a static image - no need to create mouse-over states.
   
5. Keep to the minimum the number of shades and colors used for the buttons: The participant needs to focus on the media, not the buttons, so minimize visual overload.
6. Differentiate between the low and high scores. I have a gradual shift from White (1) to Yellow (7) 
7. Make sure you have good speakers so that the participant can hear clearly the narration.

2.2 Capture any web page
The second approach makes it possible to capture user's response to any streaming content, on any site. This scenario works when:

1. You want to test content that is on a production site but you don't have the media file locally
2. You want to capture response to a section of a competitors site
   
3. You want to capture response to streaming content but are also conducting a traditional usability test for the site (navigation, workflow, tasks and so on)  
4. For some reason you can not use self-created test pages.

Just keep in mind that an internet connection will be required in the testing -- try to avoid at all cost a wireless connection and opt for an ethernet cable, if available.

How it Works:
Since a measurement bar graphic cannot be used, I suggest a low tech solution - drafting tape. The simplest method: Apply a strip of drafting tape directly to the monitor, above the clip you want to test. With a sharpie, write 'Content' in the top-center, the number 1 on the left, 2 in the middle and 3 on the right. Apply a second strip on the bottom of the clip, write 'Engagement' and the 3 numbers.

The strips help guide the user to well defined area of the screen where you want them to click. The strip is semi-transparent, so the user can see the mouse pointer then they click, and since they click in areas that are not part of the content object, the streaming is not interrupted by the clicks. When you view the recorded session later, the drafting tape strips will obviously not be there, but since you know their meaning - the clusters of clicks on top and bottom of the clip, and to the left, middle and right - will help you collect the relevant data as effectively as if there was a graphic there. Once this section of the study is done, you can peel the tape off the screen an move on to another topic.

While the example above works best for a 3 rating system, you can setup a more granular system used the left and right sides of the box. However, keep in mind that you want to keep it simple, and that adding too much tape around the clip may mask too much of the screen. Also think about the accuracy when logging the

What you need:

1. A 1" 3M™ Scotch® 230 Drafting Tape - this tape sticks to the screen but is easy to peel off.  You can get it in any office supply store.
   
2. Ultra or extra fine tip Sharpies - I use a Blue for the content strip , Black for engagement strip and red for the numbers. (Avoid using Red and Green for labels because they carry an inherent association for bad (Red) and Good (Green), which may confuse the user.
3. Small scissors (to cut the tape nicely)
4. Lens cleaner solution to wipe the screen after peeling off the tape

Keep in mind

1. Don't be sloppy: Cut the strips with scissors. If you have to tear the tape, fold about 1/2" on each side to give the strip edges and straight edge.
2. Apply the tape as horizontally as you can (Leveler is not needed...)
3. Demonstrate to the user how you want them to act during the recording and make sure they are comfortable with the mouse going 'under' the tape while they click it.

3. What's next?
Once you capture and tag the sessions, it is possible to translate data to valuable information. There are many interesting ways to slice and dice the data, well beyond the scope of this document. However, as you can see in the graphs below, aggregation of session data make a compelling story about response to content and level of engagement to existing or proposed streaming media. makes helps present to stakeholders important analysis and help develop strategies

------------------------------------------------------------------------
* Can be used with Morae 2 and 3.

The Analog Threat

Google's new browser has a privacy mode called 'Incognito'. In this mode, sites open in a new window and do not "...appear in your browser history or search history, and they won't leave other traces, like cookies, on your computer after you close the incognito window."

Google warns users that Going incognito doesn't affect the behavior of other people, servers, or software. Be wary of:

• Websites that collect or share information about you
• Internet service providers or employers that track the pages you visit
• Malicious software that tracks your keystrokes in exchange for free smileys
• Surveillance by secret agents

These are all sophisticated electronic transgression methods and they sharply contrast the last point:

• People standing behind you

This point is needed, perhapse, because, to quote Voltaire "Common sense in not so common". Interstingly, this is the only threat most users CAN do something about if they pay attention...

SOA Predictions for 2008

Dr. Jerry Smith just published his SOA predictions for 2008. Certainly, the domain of user experience design must become an integral part of SOA. Perhaps at some large companies this synthesis already takes place. But it appears that there is still a massive gap between the front and back ends. Given that simulation tools are just emerging (See Axure), and Visio wireframes still dominate the UEA practice, I doubt that a lot will change in 2008 - The gap in maturity is just too wide.

Still, I would like to append a question to Jerry's predictions:

Will companies finally realize that SOA must also incorporate UEA (User Experience Architecture)? Let's revisit next year.