During a presentation I gave to the Search track of the IntraTeam event in Copenhagen last month I caused some consternation by suggesting that Best Bets for a search application were a well meaning, but totally misguided approach to improving search performance.
Why do companies invest time and effort for less than ideal results?
Cracks in the Methodology
The initial list of Best Bet query candidates is often taken from the top 100 (or more) search queries. The Search Manager has to then find people who will know what the definitive document is to match the term -- sure to guarantee 100 percent precision -- and then not only add the document to the top of the results page, but ensure that the Best Bet remains the Best Bet in the future.
The first question to ask is whether the search logs are reliable. Often the terms are listed as single words --"bearing" -- even though the query may have been "track runner bearings." Assuming that the search logs are being used effectively, the second question is why the information is not more readily available on the intranet or similar applications.
If information on bearings is organized hierarchically, users will be led through a route which progressively refines their requirements and -- when they reach a possible document -- take them to a landing page with context and related links. A Best Bet will not have this context and will rarely include URL that enables a user to work back through related content.
One justification I often hear for the introduction of Best Bets is that users cannot find the information they need on the intranet because the information architecture (IA) is fast becoming a total mess. Search and browse are -- with monitor -- three different and related ways of finding information and it is important to find a balance between all three. Search will never offset poor IA.
The current focus on filters and facets as a way of helping users refine their searches means that little attention is being paid to providing enhanced query management along the lines of "Did you mean X?" This information will be generated by the index file and so can be displayed to the user in near real time as they start to add keywords. The only downside to this approach is that the index file will almost certainly not be controlled for permissions, which might result in query terms being displayed which could be embarrassing.