Comments on Exploring Content Filters

Paula Thornton — Thu, 10 May 2007 19:20:32 GMT

There’s the simple and then there’s the really complex. The example I always turn to is the Texas Instruments parametric search (powered by Endeca). The search begins once you’ve clicked on a product group (the visual design is not optimal): http://focus.ti.com/tihome/docs/paramsearches.tsp, the data is fully displayed at the bottom, if you scroll down and start selecting parameters you can see how the dynamic relation between the parameters affect the results.

What’s interesting about TI is that they don’t drive their main site search with Endeca, just the parts search—but TI has a great understanding of the need for designing a variety of specialized searches (http://focus.ti.com/tihome/docs/allsearches.tsp?DCMP=TIHo…, not that all of these have been ‘optimized’). A great piece on searching in general (http://www.intranetjournal.com/features/cadis-2.shtml) notes semiconductor catalogs as one high-profile use of parametric searches. The real issue is LOTS of data/attributes to sort through to get to what’s relevant to you. If the kind of content that Google returns had more parameters in the metadata then we’d see a lot more parametric searching going on. Otherwise it’s a must-have where there are lots of attributes and large data sets (Home Depot’s products are navigated via this approach, again using Endeca).

Gabe — Tue, 28 Aug 2007 00:28:31 GMT

An excellent example of this kind of filtering is on http://www.epinions.com/ – the site’s a great hybrid of exclusive category navigation and filtering (“I want cameras that cost less than $300 and are made by Canon”)

Clifton — Fri, 26 Jan 2007 14:50:13 GMT

A few months have now passed and I’ve found a great example of this technique used online at Oddbins.

—- This is an excellent Filtering example!—-
http://www.oddbins.com/products/Wine.asp

I have also linked the demo that was mentioned above as well as an example from NBA.com that almost ‘gets it’... but not quite. :(

—- This is the direct link to the FacetMap demo—-
http://facetmap.com:8080/browse.jsp?s=220000&n=2&v=2

—- This example is almost there… Such a shame—-
http://store.nba.com/

Why is it that the only perfect examples of Filtering are about drinking? How strange is that?

I guess you have to be drunk to warrant this amount of ‘ease of use’...

Andrew Otwell — Tue, 28 Aug 2007 00:28:31 GMT

Small addition to my comment above:

The URL for the FacetMap demo is better written as http://facetmap.com

Travis has done added some impressive things here, including the ability to try out different navigation interfaces for the facetmaps, including a drop-down menu system almost *exactly* like what Clifton is describing in his article.

Andrew Otwell — Tue, 28 Aug 2007 00:28:31 GMT

It does sound a lot like faceted searches. You should check out the stuff that Marti Hearst has done at Berkeley; I think she’s explored some of the concepts here pretty rigorously. Really I think the word that’s missing here is “winnowing”: a method of refining a large set of stuff into a small and precise set of stuff.

Also look at the examples at http://64.220.144.86:8080/facetmap/index.html where you can look at an example using wines or create your own facetmap to try it out. (when did that cool feature get added?)

For the “empy node” problem: we’re building a faceted search interface for a large product database right now and decided that any nodes with 0 results in them would simply not be shown as navigation options. (They of course appear in the CMS for content creators, and would appear with results if content is put in them).

I’m wondering about the real value of this for small screens, as you suggest. Part of the power of it is taking advantage of people’s ability to scan large bodies of information quickly and to decide how to process it. Without being able to actually *see* the large set of items being refined, winnowed, or filtered, I don’t really have any idea about what I’m getting.

Avi Rappoport — Tue, 28 Aug 2007 00:28:31 GMT

This seems very much like faceted browsing to me. Peter Moreville has written some stuff about it. The problems I’ve found with faceted browsing include a lack of instant gratification (search engines give you immediate results), and issues of context, especially feedback on what I’m doing.

Marti Hearst at UC Berkeley has some good examples at her Flamenco project (see flamenco.berkeley.edu). Marti’s got an extra improvement: she makes sure that every option shows the number of hits you’ll get if you choose it. This addresses the “empty nodes” problem as well as the “one item surprise”.

lee fleming — Tue, 28 Aug 2007 00:28:31 GMT

This is not exactly new. We’ve used this technique for years, even using the choices to filter the choices. Works great. Customers complained a lot before we used it (who wants to wade through long, long lists of anything other than one’s personal stock holdings?) They don’t complain now – I guess that means it’s usable.

PeterV — Tue, 28 Aug 2007 00:28:31 GMT

Interesting. We actually tried out an approach like this for a client – no usability testing though so no idea if it worked well.

There are a few problems with this approach we encountered:

- empty nodes: you have to have your dropdowns dynamically avoid empty nodes.
- if they have to choose all three dropdowns before clicking submit, that can get complex cognitively. It may be better to reload upon choosing an element in a dropdown, but that invites some obvious interface problems.