Leanord Richardson has cooked up the Ultra Gleeper essentially a single person, or small group, recommendation engine for Web pages. The Ultra Gleeper is interesting in that it uses blogrolls, watch engines (Technorati, del.icio.us webfeeds), and ratings in an attempt to avoid a number of standard problems with recommender systems. You can read all about the design in Richardson's CodeCon paper.
I like it a lot as a potential experimentation platform, but in thinking a bit about potential uses, I forced myself into a conundrum. Without a specific task, what do people need recommendations for? This is a case where good enough is the enemy of best. If I'm just browsing, surfing, and trawling the blogosphere, I've got more than enough feeds to provide good morsels. If I need more serendipity I can just tap into Findory or the various dexes.
But if I don't have a particular task, how can a technology say it's made my life any better? Greg Linden will say news personalization has "learned what I need" and is giving it to me, but by what metric?
Alternatively, how do we know the Ultra Gleeper or Findory are doing any better than random? And what does "doing better" mean?
However, all hope is not lost. There are plenty of places where a task might be explicit or implicit and a recommendation engine might be appropriate. For example, if I really am a blogger, writer, analyzer, synthesizer ( no jokes from the peanut gallery sometimes I wonder too), then a recommendation engine can keep its eyes on the lookout for corners of the world I should be interested in, given what I write on. This is a soft definition of what I need, but at least there's some hope of knowing that I get better at it by getting what I need. And there's plenty of contexts and activities where a similar story can be told. I pity the recommender's though, who don't have much of a picture of what the user is trying to achieve, if anything.