Jeremy Pickens, Gene Golovchinsky, Chirag Shah, Pernilla Qvarfordt, and Maribeth Back, all from FX Palo Alto Laboratory won SIGIR paper of the year 2008 with “Algorithmic Mediation for Collaborative Exploratory Search“.
It’s a really interesting paper. It’s about searching in a completely different way than we do at the moment. The idea is that users with the same information need team up and search together at the same time. They’re provided with tools to help them search and collaborate, and also “algorithmically-mediated retrieval to focus, enhance and augment the team’s search and communication activities.”
Recomender systems can use the user profile to reference certain characteristics, ask the user to get involved and vote on things that are useful or not for example, rank items, and so on.
Collaborative filtering systems have agents that collaborate to find information and filters that filter patterns to make sense of it. They will look for users who share similarities with a profile, and use their data to make predictions. Amazon was one of the first to propose rating products way back in 2000 to add to its variables. Another types of collaborative filtering will look at the user activity history to make predictions.
Sites and services that use this kind of method are already very popular. I regularly use Amazon, Digg, Stumbleupon, iTunes, iLike, Last.fm and a whole host of others.
I’m not sure however if I would like to use this kind of thing for searching the Google index for example. In my experience the recommender systems and collaborative filtering systems are very useful and I discovered all sorts of new music that I like this way, and books I’ve enjoyed reading, but there are also a mountain of suggestions that are not relevant to me.
I use search engines all day for work, and I don’t have time to be proposed something I might find more interesting, having said that, how accurate are my results anyway? Well I always manage to find what I’m looking for one way or another, so it can’t be too bad. But how much better could it be?
Search engines have trouble with vague or ambiguous queries and currently the preferred solution seems to be to apply personalisation or query context. I don’t think personalisation has been tested to its full potential just yet and I think it has the potential to improve things. Query context information is useful but the problem of “query drift” (user changing his/her intent) still remains.