Susan Dumais from MSR is our first presenter today, and explicitly released what she's showing from the NDA. Yay!
She's showing some really nifty stuff, including a personalized search tool that lets you do a web search, then drag a slider to make the results more customized based on what your local computer knows about you. It's a split screen result, so you see the original results on the left, and the increasingly personalized results on the right. Very, very cool.
I spoke with her last night about some of my feeling that what I want from search is to be able to find things that are important, which for me means being alerted if things that are typically similar suddenly diverge, or if things that are typically very different suddenly overlap. (I wrote about this on M2M some time ago, in a piece called The Power of Overlap.)
Susan is followed by Eric Brill. He's talking about how it's often difficult for users to extract the relevant information/answer from a larger document in search results. Information-centric search" rather than "document-centric search." He shows some "AnswerBot" technologies that let you ask a question, and be provided with a range of suggested answers, complete with probabilities. He shows an example of the question "Who did Britney Spears run off and marry in Las Vegas?" and the suggested answers of Jason Alexander and Kevin Federline ranked first and second. Clicking on a suggested answer provides the supporting documents.
He then talks about how this works on the back end, in terms of the AI of parsing both the user's question and the possible answers. Turns out that the size of the web, and the repetition of content, makes it much easier to locate patterns that are likely to answer the question. Great line: Moving from: 'Does the page contain the query terms' to 'Does the page satisfy the information need'
Lili Cheng from the social computing group is the last up, and she shows some of their "personal map" work, where your activities (email, calendar, etc) affect the way your contacts are arranged and grouped. But, she points out, we already know who we interact with. How can we put this in a larger context. For example, who do you know that knows someone else--basically the LinkedIn facilitated introduction idea. But again, what's interestingn to me is how this shows overlap. It lets you map how you're connected to another person, and through what paths. I'm less interested in the endpoints than the nature of the paths that lead there.
She shows and talks about Wallop--of the 424 people invited in, about a quarter have been regularly active. What this lets them do is build maps of inferred social networks--"who's important to whom." People want to explicitly control, to be able to add/remove people. But they also don't want to have to spend a lot of time organizing. They're going to start to do a larger scale deployment of Wallop, seeding the network with controlled invites (please do not leaave comments here asking me how to get an invite. i cannot get you an invite. really.).
So, how can this be integrated into other activities, including search. Apply it to email, for example--put the people most important in your current network in an easy list, click on their name to get all the communication with them. This idea of integrating the social tools into other tools is great!
(What we're seeing supports my sense that MSR has some amazingly smart and interesting people working on these problems. Why does this keep surprising me? Why is there such a disjoint between what they produce and these smart people who are helping to produce it?)