I was speaking with a client about faceted search. (For more information on that subject see my blog post “A Firm Foundation for Faceted Search”). We discussed the need for well-organized and well-structured data to support useful faceted searching. The client challenged that need, and stated she had read and been told that some forms of guided navigation require no data preparation—and will work with completely unstructured data. In this post I’ll compare and contrast two approaches to searching.

I pointed out that search clustering is indeed an alternative to faceted searching, especially for unstructured data, and while it too is a form of guided navigation, there are important differences between the two. Let’s take a closer look.

Pay Me First

I often tell our customers that faceted search is simply Boolean fielded search for the novice. Facets are after all just fields, but whereas most novice users are never going to use the “advanced search” page with multiple fielded search, they love to construct complex queries by adding one filter (facet) at a time. However, a superior faceted searching experience requires work. It needs well-catalogued data (e.g. the use of controlled vocabularies to populate fields/facets), thereby enabling the clustering of the search results based on a specific field. So faceted search is a “pay-me-first” model: do the work upfront and you’ll get a good search experience.

Pay Me Later

Search clustering, on the other hand, is a “pay-me-later” model. Search clustering requires no upfront work on the part of the content manager, but rather uses linguistic algorithms to cluster (or classify) search results on-the-fly into clusters (related content) that the user can then browse in order to filter and narrow the search results.

So why would (and when should) you choose faceted navigation versus search clustering?

Straight Up or Blended?

In real-world applications with structured data and high-quality metadata, faceted search dominates. This is because the facets are domain-specific. For example, the facets for “Country” will be a list of countries, the facets for “Time Period” will be a list of years or time ranges, and the facets for “Subject” will be a description of the item. This means users get a highly precise and understandable way to filter and narrow search results. But with search clustering of unstructured data, the “Topics” displayed will mix and match some of these domains—sometimes the topics are descriptions, sometimes the topics are geographies, sometimes they might be dates, etc.—so your ability to filter search results with precision is impaired. This is why almost all ecommerce sites use faceted search versus search clustering.

When shopping for Probiotics the user wants to filter by facets such as “Format” (e.g. pill or powder), by the “Dietary Specialty” (e.g. Gluten Free or Organic), and by shipping cost (e.g. free or not). The user does not really want to navigate topic clusters such as Gut Microbiota, Dietary Supplements, Immune Response, Irritable Bowel Syndrome, etc.

Perception and Reality

Another downside of search clustering is that because the clustering is done in real time, it can appear to be slow in a world where we have been trained by Google that all search results are instantaneous. In addition, for larger search result sets, search clustering engines often only cluster the top results in order to make search seem faster.

So, if your data is well-structured and well-organized, add faceted searching to your website. Your users will love you! However, if you have mainly unorganized and unstructured documents, you may find search clustering is a big help when added to your search interface.

Do you have experience with search clustering or faceted search? If so, please share your observations in the comments below.