« Launch of the global Intranet Innovation Awards | Main | Presentation: IA Summit 2007 »

Unanswered search questions: areas for further study

Yesterday I ran a half-day workshop on designing search at the IA Summit in Las Vegas, and today I listened to a talk on search analytics by Lou Rosenfeld and Rich Wiggins.

We are all saying the same things. Namely, to make search work a lot better:

• track most popular searches and failed searches
• check the popular searches yourself, and fix the results if required
• use search engine synonyms
• implement search engine 'best bets'

These are all practical approaches that are easy to implement and they will dramatically improve the effectiveness of search results. (See my earlier article for more on this.)

At the end of the day, these are all very simple (even simplistic) approaches. They cover the basics, but there is still a lot of work (and thinking) to be done before we have really cracked search.

What's become apparent to me is that beyond the basics, we quickly run short of experience and answers, across the whole search industry. There are few practical projects to draw experience from, and what results there are haven't been widely or efficiently shared with others.

So in the spirit of encouraging further research and discussion, here are my big unanswered questions of search (in no particular order):

• How do we understand the context of users' searches? Depending on the context, users searching for 'disposal' may mean completely different things (are they looking for recordkeeping information or garbage bins). These are also called 'ambiguous terms', and they are a real problem to be overcome if search is to work really well. How do we solve this problem?
• How do we blend search and browse? Search vendors offer an incredibly rich set of tools that can be included on search and results pages for filtering, navigating, and refining. When should these tools be used, and how should they be presented?
• How do we process and cluster search usage reports? There are different possible search terms for the same search, but at present the "most popular searches" report simply adds up the number of times the same (exact) search string is used. We need better algorithms for processing, cleaning up and clustering search usage reports, to make them more useful.
• How do we more intelligently match 'best bets'? As above, 'best bets' are typically matched very simply, with a specific search string giving a pre-selected list of matches. As usage of this grows, we often end up not giving any best bets, or presenting too many. What are the algorithms for better matching best bets?
• How should we display best bets? Should we present best bets in its own section at the start of the results, or just as the first few hits?
• How effective are automated clustering tools? Search engines now offer many automated ways of analysing text, providing recommend links, terms and clusters. How well do they work, when should they be used, and how should they be presented?
• How do we create a dialogue between users and search? We know that users often don't get what they want with their first search, but it's proving very difficult to get users to then "work with" the search engine to obtain better results. We know it's not Ms Dewey, so how do we create this dialogue?
• Can we use social software to improve search? Google works very well because of "pagerank", using the cross-linking between sites to improve search. On our sites, and within the enterprise, we don't have this information. Will web 2.0, enterprise 2.0 or social software give us this valuable context to help with search, and if so, how?
• How do we make enterprise search work? While vendors are madly selling enterprise search products at present, the reality is that few (if any) know how to make this work in practice. The fundamental challenge is: how do we determine relevance and give meaningful results when searching across such a wide range of information and formats?
• Are their other good search interfaces, beyond just a search box? There is some experimentation now with a variety of different search interfaces, including "speedsearch" that displays possible search terms while the user is still typing. Which of these new ideas work, and when should we be using them?
• How do we best meet the needs of specialist search users? Specialist search users (lawyers, engineers, researchers, etc) have very different needs from general searchers. Are there "best practices" about how to best design search for these users?

If nothing else, this list may help people to say: "hey, the nifty solution I've been told to implement is on this list, maybe there's some hard thinking to be done". The questions can also be a good foil for vendor over-enthusiasm and feature-obsession...

Note that this is just a list written off the top of my head, and I'm sure I'll think of a dozen more questions before the end of the day. Email me with your big questions, and I'll synthesise all of them into a single list.

Posted by jamesr on March 25, 2007 08:18 AM
Categories: Search tools