A hot new topic at this year's Enterprise Search conference in New York was faceted search, and how it compares tithe use of parametric search and taxonomies.

In a sense, faceted search is a new spin on parametric search and taxonomy-based search. We've discussed taxonomies before, and in a way faceted search is a natural outgrowth of taxonomies.

Let's look at a few examples:

• On a consumer-centric site, a search for the term “outdoor” might be filtered by a attributes such as 'apparel' or 'sports equipment'. Searches within apparel might be further sub-divided by color, style or size.
• On an employee centric site, a search for 'manager' could be presented by the attributes 'resumes' or 'harassment policies'.
• On a software vendor site, a search for specific error message could be limited to a specific range of software versions, or to a particular operating system.
• On a sales-force / CRM application, a search for 'reference account' could be targeted by region, industry, or Fortune 500 ranking.

All of these attributes are really facets.

Parametric / Faceted search takes this idea to the next level. For any search that you perform, you are pro-actively shown a summary of the number of matches for your search for specific values within each available field. These matches are hyperlinked, and by clicking on one of them, your search is re-run against that specific subset of data. Those results are displayed, along with additional other fields and attributes that have matches, that you can in turn click on. This allows you to “drill down” through results, getting more and more specific. Field values or attributes that produce no results are not shown. And any constraint can be easily removed from the search, to open it back up to more documents.

This goes beyond the old fashioned “advanced search form”, in which every available field, with every available value, is presented. Users never liked those screens anyway, and it was often possible to accidentally select a combination of values that didn't even exist in the document dataset. Faceted search replaces this mess with a much more interactive and helpful framework.

Adding faceted search to our examples above we get:

• On the consumer centric site, as you browse apparel matching the search “outdoor”, the search results has a right hand column that shows the number of matches for each size, style and color. If you click on a particular size, the results are filtered and re-displayed, now showing the number of matches in that size, and showing how many items are available for each available color. If a particular color is not available in that size, it is not offered. A customer could even “cancel” the size criteria, and choose to instead start filtering by color first.
• On an employee centric site, an initial simple search for the word 'manager' would have brought back quite a few results. The system would automatically show the number of matches for that word within the 'Resumes' and 'HR Policies' databases. By clicking the HR Policies document count, the search would be rerun just against that database. At that point, additional drill down options might also be offered.
• On the software vendor's site, an initial single word search for the word “error” would have brought back many results. A user could have then drilled down several times, through 'Support', and then on the specific product name, then perhaps on the version they have, then into 'installation notes'. Now the single word 'error' is highly targeted into a rather small set of matches, and the user has a much better chance of finding the answer. Note that the system would only present version numbers for the product the user chose; this is a key concept of faceted search.
• Our hypothetical sales-force / CRM user who is looking for a 'reference account' could quickly drill down by region, then by industry, and then by sales volume or Fortune 500 ranking. At each point they still see a subset of matching customer records; at some point a 'good fi'” will be obvious. A busy sales person is much more likely to click their way through this type of interactive drill-down presentation than they would be to fill in some blank 'advanced' search form with 20 different fields.

Parametric / Faceted versus Taxonomies

These areas certainly have a lot in common; in some respects parametric/faceted systems could be viewed as simply the next logical evolution of taxonomies.

Traditional taxonomies offered a type of 'drill-down' approach to narrowing searches, which is still popular. Taxonomies tended to be subject, product-line or org-chart based. Taxonomy branches tended to be somewhat fixed, and contained textual data. 'Facet'” tend to be dynamic in nature, and can often accommodate data types such as dates, numerical ranges and geographic coordinates. From a style standpoint, these new systems tend to presented in slicker, more dynamic user interfaces.

Mathematically speaking, taxonomies were'“trees', whereas these faceted systems are more 'graphs'. Some modern taxonomies can actually be viewed from multiple primary axes, so the same set of taxonomy nodes can be dynamically viewed in more than one tree arrangement.

If you have a taxonomy, great! They're not out of fashion - yet. And taxonomies enjoy much wider support by vendors. But you should stay up to date; perhaps you could augment your taxonomy with some of this new technology, to give users even more granular control over their search results.

Why this Parametric / Faceted Feature is Becoming a Priority

'B2C' web sites figured out the importance of parametric search early on: customers would type 'pants' in the search box and find hundreds or thousands of matches. Consumers already understand choices like 'color' and 'size', so presenting these options was an easy way to assist them in finding their quarry.

But enterprises are drowning in data as well, and users still often issue one or two word searches. Instead of the prospect of lost customers faced by B2C sites, companies are worried about the lost productivity of their knowledge workers. This is compounded implementing the seemingly laudable business requirement to 'search enable all of our corporate content under one unified search engine'. As more and more companies move towards this goal, the amount of documents that match a single word query increases almost exponentially.

Properly matched and implemented, these systems have the potential to drastically improve the search experience, by leaving humans in the drivers seat as they quickly drill down through thousands of matches.

We'll continue to extend our coversage of faceted and parametric search in the next months here in Enterprise Search.