UX matters and BI matters at new IEEE site
by Pete Bell
Pete Bell
The new IEEE Xplore Digital Library site just launched and it has an interesting type of numeric range filter, worth discussing from both a user experience and business intelligence angle. This filter is among several noteworthy faceted search designs at IEEE, including the choice to give each facet its own search box and breadcrumb.
IEEE launched just after the UX Matters blog ran its fantastic piece on numeric filters by Greg Nudelman. Greg nailed one of the most common mistakes I see on faceted search sites, which is to offer users ranges in arbitrary bands, rather than ones that map to user tasks. For example, Greg shows the negative pattern of a digital camera navigator that asks people to pick bands of how many megapixels they want — 9.0 to 9.9 MP, 10 to 10.9MP, etc., which is definitely not how people set out to shop for cameras.
In contrast, Amazon offers its ranges for customer reviews as 2 Stars & UP, 3 Stars & Up, etc. This makes sense because people would never want to see just 2 star reviews – they’d want to set 2 stars as a lower bound. (Of course, the Amazon design is also limiting — there are other useful tasks you could imagine.)
For many tasks, people want to set both upper and lower bounds — for example, by setting stops on each end of a slider. Kayak.com was one of the first sites to do this well, with sliders like “take-off time” and “layover duration.” But there are two big gotchas here. The first is that it can be easy to set a range that gives no results. Depending on the dataset, those dead ends can range from annoying to prohibitively annoying.
On his recent UIE virtual seminar on Search & Discovery Patterns, Endeca’s Mark Burrell showed a solution to this from the Endeca pattern library: display a histogram of result counts above the slider. Greg ran with this pattern to propose this great re-design of price sliders at Trip Advisor, which he likens to an Edward Tufte sparkline, an “intense, simple, word-sized graphic”:
Which brings me to the IEEE site, which has a different version – in this case, for timelines:
That slider corresponds to a search for “Robots, Locomotion.” (Type-ahead assisted me in getting that precise-sounding search term.) As you can see, they grey-out years for which there are no documents in the collection. And here we get much more than a guardrail to prevent us from selecting dead-end ranges. It really is a sparkline that tells me a short story about the history of robot locomotion. I see the year when the first research appeared, some sputtering, then an uptick in frequency over time. In fact, this is a good timeline of most technological innovation. (I think a full histogram of counts per year would fill out the story even more.)
And that leads me to the second interesting angle on this topic, and the second gotcha to implementing range filters with histograms. This range filter is really a form of business intelligence. Adam has discussed in this blog a bit about where search and BI converge, and IEEE is a material example. When the user is focused on the histogram of the publication years, he might be more interested in learning about the set of results than in any single member of the set. And BI has mature visualizations that could help on tasks of this kind, like heatmaps, charts, etc.
To be fair, most IEEE users will be more interested in the timeline to help them refine down to a single result. But it’s not hard to get from here to other content collections where users are interested in those set-based views as often or even more often than in single results. In fact, I think IEEE is doing some very cool text analytics on its collection, which could lead to some new types of visualizations.
And the second gotcha: in the UX Matters blog, reader Harry Brignull commented “Your sparkline / slider idea is a good one, but it would be computationally expensive. I don’t know offhand any high traffic sites that use it. Maybe Endeca have something up their sleeves / coming soon?” Daniel Tunkelang pointed out some potential computational shortcuts in the comments, like approximation. Still, in this use case, we need to know precisely which years have a zero count. Depending on the scale of the collection and how much BI we want to layer in, mixing search and BI can very quickly get us beyond the territory of UX Matters into the world of backend architecture matters.
on February 18, 2010 at 4:31 pm
Permalink
Thank you, Pete — I am glad you found my article useful! Fantastic example of the timeline sparkline on the IEEE site. It proves nicely Daniel’s point about approximating the sparkline using computational shortcuts.
Greg
Continue the conversation on Twitter @DesignCaffeine?