Currently, third on the list is poorly implemented faceted navigation. It is fast becoming the fix-all for search applications where users do more than skim the first 10 results from a list of 50,000. At one time, a "Best Bet" was the preferred solution until it became obvious that no one could agree on what it was. Now, people who should know better (which includes most search vendors) are offering some flavor of faceted navigation as the ultimate panacea to information overload.
This short polemic was catalyzed by a consulting gig where I found myself looking at the search screen of one of those magical applications. The search term I typed in was one of the technical innovations that would be right at the center stage for this organization's future success. I was not surprised to find more than 4,500 results (earlier I had found 119,000 using "confidential" as the search term). To help me sort these, the left-hand side of the screen proffered a display that told me there were 3,200 results in the intranet, 600 in the document management system, and three in some specialized application with an acronym as a descriptor. Now, am I alone in finding that to be of no value at all in helping the user decide where next to search?
Another facet-based approach is to tell the overwhelmed searcher that there are 2,300 PowerPoint documents containing the keyword, 2,200 PDF files, 1,500 Word documents, and 357 Excel documents. This really does not help. If searching is like looking for birthday presents in Macy's, this is akin to being told that there are 5,230 presents in Home and Kitchen-not terribly useful.Before any organization turns to faceted navigation as the path to user satisfaction, I would strongly recommend reading Faceted Search (http://thenoisychannel.com) by Daniel Tunkelang, formerly chief scientist of Endeca and now technical lead/manager at Google. Tunkelang provides a very concise account of the development and practice of faceted search, both from the back end and the user interface. Reviewing his post and reading Marti Hearst's book, Search User Interfaces, reinforces how important it is to take a user perspective on this technology, which has its origins in the work on a Colon Classification developed by S.R. Ranganathan in 1933.
There is a larger issue here. Only rarely do I come across organizations that have done serious usability testing of their search interfaces based around work to determine not only types of questions that users are asking but also why they are asking them and, above all, what they are expecting from the search engine. Recently, the Royal Society of Chemistry (www.rsc.org) redeveloped its website with faceted navigation to over 500,000 journal articles. Extensive user research over several months resulted in a search interface which is optimised for research chemists, who are the primary users of this site. It takes that level of effort to get the best from faceted navigation.
Of course, even the most rigorous implementation procedures fail to cope with changing user demands. Returning to the first faceted example I gave in this article, I should note that, at that time, the technical innovation I used as my search term was one that was still new for the organization. However, that makes it even more important that those seeking information on this emerging technology within the organization be able to find not only some results but all of the results on any specific aspect of that technology.
To achieve this, the search team must work through the search logs, talk to experts, take some of the core documents and add in metadata to drive some specific (useful) facets, and test the results with users. Working out what might be the best facets for users has to start right at the outset of the decision to purchase, and not when it goes live. Unfortunately, too many organizations seem willing to spend perhaps more than $1 million on the search engine but fail to invest in the creation of a search team. Ignoring this critical success factor for successful search implementation will inevitably yield results that fail to meet user expectations.