Enterprise search still a technology conversation

In short, best bets (where an editor can select the top results for certain search queries) is seen by many information professionals as about the cheapest and best way to improve your search engine, but the enterprise search industry doesn't have much of a clue. Many enterprise search products don't explicitly support this. More generalized, most companies seem to think of search as a technology problem, whereas most of the consultants and experts understand the importance of adding people to the mix. In 2003, I started an article at Onlamp like this: "A useful search engine is more than a search algorithm. This article explains how to create a search query analysis tool, a best bets feature, and a basic controlled vocabulary." The idea was to write for the techies who are building the tools about what we, information architects, think are the things missing from most search engines. Onlamp is O'Reilly's publication for open source hackers, and I was on a mission to spread the word about IA (also to other groups, like designers). My point was: there are easy things you can add to your search engine that let humans add value to it, like best bets, or a search log analysis tool. It's not rocket science - if I could write a techie how-to article, the search vendors should be able to figure this out. Last week, at the 2005 Enterprise search summit, I did a little unscientific survey with the vendors about best bets. I asked them if they had such a functionality in their product (I had to explain it to most), and what they called it. The results were in line with my overall impression of enterprise search. Most of the products work like this:

Spider content and rank
Auto-generate and auto-populate taxonomies to add value to search

Notice the absence of humans in that process. The control panels of the products tend to contain a section with sysadmin-like functionality, and some analytics (most allow you to see what search queries people have been using). Most of them assume that the person using it has been trained to use this tool. There is surprisingly little functionality aimed at the person whose job it might be to tune the engine with best bets and such. The people I spoke with who actually do that job, use things like Perl scripts or open source software to analyse search queries. (For example, I was told Googlebox doesn't handle logging multilingual search queries (it searches fine), so one person used Webalizer instead.) When I asked the best bets question ("does your product do best bets, defined as ..."), even after explaining the functionality, I got surprisingly many blank stares. Best whats? Why would you want to do that? Some products have best bets, but the closest a lot of them could come was to say you could create rules to improve the result of certain documents. That's like saying, sure, you can do HTML with Word. In theory perhaps, but it's not really useful. Here is an incomplete list of products that do best bets, and what they call it. This is an unscientific and uncomplete survey, which may have mistakes in it. Don't use it to judge a particular product, use it to get a sense of the field.

Autonomy: you can kinda do them through rules.
BA-insight: ~~no best bets~~ Yes, through SharePoint.
FAST: yes (although I have doubts here).
IBM: yes, they're called Quick Links.
ISYS: not really.
Mondosoft: yes, they're called Top Hits.
Open Text: it's coming up in their next release.
SER Solutions: no.
Verity: yes, calls them Sponsored Links.
Vivisimo: yes, kind of.

I didn't have time to ask the other vendors - feel free to add in the comments. By the way, to work well with users, best bets should appear in-line with the other search results, not separate from them. If I was to do a more complete survey, I'd add that in as a criteria, together with an easy to use admin interface, CV functionality and an easy to use search analysis tool that includes analysis of suddenly popular queries.

# May 20, 2005

← All posts