ia play

One of my great hopes for our current intranet project is to significantly improve the intranet search.  The current set-up used the search bundled with Stellent. It is universally derided within the organisation and with good reason (the Stellent search itself may not be at fault, I imagine some changes to the configuration could fix some of the more significant problems).

I’ve heard mixed reports of Sharepoint search. Our suppliers are very positive about it, and it does seem hard to imagine how it could be worse that what we currently have.

At the TFPL conference I attended Sharon Richardson of Joining Dots defended SharePoint search. She went a bit far with the statement “…so the problem with search is not the technology, it’s the users” but there’s some interesting stuff in the ‘research‘ she referred to.

55% The content was badly named, didn’t contain the words the users was searching for, wasn’t easily identifiable in search results (e.g. if you have 2 results both called Cafe – which is for London and which is for Manchester?)

30% The content users were looking for didn’t exist

10% Users were using wide or strange search terms (why would somebody search for ‘google’ on the intranet? what exactly did they want to find when they searched for ‘form’?)

5% Search wasn’t finding appropriate content or ranking wasn’t appropriate

I’ve been keeping track of failed or problematic searches on our current intranet. Not particularly scientific but it has been an interesting starting point for evaluating the new search.

30% mismatches in language
25% inappropriate date ordering
15% lack of stemming
15% overly rigid phrase order matching
10% ambiguous queries
5% inappropriate alphabetic ordering of results

If a number of results are assigned the same relevancy then they are returned in date order, and if there are a number of results published on the same day then they are returned in alphabetical order. The relevancy scores don’t seem to distinguish between enough results, so the date and alpha ordering are regularly skewing the results.

The mismatched language and the ambiguous queries are sure to still be problems with the new search. I’m not going to endeavour to ‘fix the users’ here. There are plenty of solutions (best bets, related searches, faceted filters and synonym control) that we can utilise.

Interestingly my experiences with our existing search have suggested that searching for just ‘form’ can be an intelligent, considered tactic in less than ideal circumstances. If you are looking for the sickness form but you are not sure if it is actually called that (absence form, sick form etc) then searching for form and scanning the results can be your least worst option. Given our current search is pedantic in it’s insistence on exact phrase order, I find myself conducting single word searches far more often than usual.

Related posts
SharePoint search: Inside the Index book ‘review’
SharePoint search: some ranking factors