Dan Rose and Jan Pedersen from Yahoo gave a talk at UC Berkeley on "How Search Engines work; Usability and Search" (PDF).
The fourth slide is particularly interesting to me. It says search is an iterative process, a back-and-forth where a person asks for some information, gets something back, refines the query, and so on.
Current search engines don't behave this way. If you search for "personalized news" and then refine the query to "news personalization", the two queries are treated as independent. The information in my history, about what I just found or failed to find, is ignored.
This needs to change. Search engines need to pay attention to what I just did and help me track down what I need.
The rest of the talk is also excellent, a useful survey of what users do when searching. It's clear that users already don't understand the tools available to them, boolean keyword search. This might give pause to those who claim that people just need more powerful search tools, seeing as people don't use or understand the tools they already have.
Thanks to Danny Sullivan for pointing to the SIMS 141 lectures and Professor Marti Hearst for hosting the talks and making the slides available.
I'm looking forward to seeing the next presentations, especially the upcoming talk from Peter Norvig and Sepandar Kamvar from Google.
Update Videos and copies of the slides for most of the SIMS 141 talks are now available. The talks include people such as Susan Dumais, Dan Rose, Jan Pedersen, Marc Najork, and Sergey Brin. Unfortunately, the Sep Kamvar talk was not recorded, darn.
Update: There are also better quality versions of a few of the talks on Google Video.