Clarify and Refine

At Endeca Discover I had an great conversation with Daniel Tunkelang about his idea that, with respect to guided navigation, we should distinguish between clarify and refine. First, we must clarify the meaning or context. Are we in the right ballpark regarding the searcher's intent? Clarify is all about disambiguation.

Then, we're ready to refine or narrow. Exactly what type of widget do you want? Refine is about increasing specificity. In Is Search Broken? Daniel hints at how we might cut through facets to clarify by leveraging guided summarization.

It's a subtle distinction, but from a designer's perspective, I think it's a valuable way to frame the search process. Clarify, then refine. What do you think?

Hi Peter,
I'm working for Kelkoo, a European comparison shopping engine where guided navigation is used heavily ;) So yes, the "clarify then refine" pattern is one of our key concerns in terms of usability, even if we have not conceptualized this concern as clearly as you did :)

I had a look at the slides and the "guided summarization" concept, but I'm not sure I fully understand it with the given examples. Could you kindly give details in your own words?

Thanks for this interesting article,

Sure. You'll have to look at the original (the .pps link in the text) to see the before/after versions. In SlideShare, see slide 45.

In this particular example, the user has searched for "storage". Rather than simply present the full set of facets for refinement, clarify asks the user to disambiguate the query. Did you mean toy storage or computer storage?

This is just one example. I think Yahoo's two-step approach to Search Assist...

...also fits the clarify, then refine model.

Daniel has posted a response...

...that further explains clarify/refine.

Hi Peter,
Thanks for the links!

We did this a few years ago for AOL Shopping, the version I worked on is long gone of course.

I wasn't involved down to the Algorithmic level but the general idea was:

Based on the number of results returned and the number of different categories, either return results to help disambiguate the term or filters to refine the search. Of course you also try to disambiguate the term during entry, but so many users ignore it at that step you are going to need to do it after as well.

It was much like Yahoo!'s Search Assist, except it had a softer tone, and it was several years ago.

One of the problems with disambiguation is it is really hard to do well algorithmically. It relies on a lot of data to produce results that exclude absurd examples; this is where being bigger is better.

It's good to see Endeca thinking about this, I know they've spent years refining the refinement part of their solution.