5th Dutch-Belgian Information Retrieval Workshop (DIR05) 10 and 11 January, 2005 De Uithof, Utrecht University, Utrecht, the Netherlands http://www.cs.uu.nl/dir2005/ Trip Report by Lloyd Rutledge I attended the first day of this workshop. The opening keynote was by Mounia Lalmas of Queens University, whom Jacco and I met at the Dagstuhl Semantic Web workshop in 2002. At Dagstuhl and DIR, she discussed her INEX work, which is an evaluation of XML information retrieval. This means using queries that account for XML-defined structure as well as content. Our buddy Arjen is also involved in this, and she and Mounia have exchanged "100s" of emails about it, along with a dozen others in the project. When I mentioned the Semantic Web workshop, Mounia winced, giving the typical remark that she mistrusts the Semantic Web because it makes such grandiose claims. (Kind of hard to sell your car if everyone has heard you have a helicopter.) In general it seems the IR people feel you get lots of emergent semantics from bulk automation that is as good as or, often, better than human-crafted semantics. You may get half the quality, but it's for free. Plus, they feel the Semantic Web hasn't really demonstrated anything to them yet. At the end of her keynote, Mounia admitted wincing at users as well. This is in the context of this year's start on acknowledging the user in INEX work. My end-of-talk question was, bothersome enough, about users: since Google-esque search uses much subsequent browsing in context, how about the possibilities that structure-based search raises? If search returns give multiple components within the same document, could the user best use a link to a higher level overriding section than to the separate subsections, the encouraging navigation to the probably relevant siblings as well? She replied that these decisions are already relevant anyway: the question of which section level is already wrestled with, even without the user. The user may tilt the question in one direction or the other. During coffee Mounia said she hopes to leave XML and go back to her original research concern: more general acknowledgment of structure around search components. This workshop is slowly getting increasingly international. Most papers, of course, were Dutch(-Belgian). UvA's Maarten de Rijke's name appears on three paper. Jaap van der Herik and Theo Huibers also have papers from their students. UU's Roelof van Zwol hosted. I borrelled it up with Loes Braun and Floris Wiesman from Maastricht (Jaap van der Herik was not there). Loes's paper, as with the Groningen ToKeN2000 workshop, was on help physicians find articles relevant to patient's needs. Having bothered Mounia with users, I bothered Loes with semantics (though mercifully off-line during the borrel). Her use case of getting information relevant for given patients (ie to prevent harmful cross-medication) seems a better application for semantics than for automated extraction. While I waited until the borrel for my question, someone from the audience did comment on how dependable quality of information is vital for doctors, and having Google-level automated quality won't cut it for doctors for long. On the other hand, if patients have data fields, it's best to expert system these into inferences that warn of bad cross-medications. This involves more data fields than you can usually cram into a text query, and even if you could, you're getting back a list of articles instead of a single statement: don't use this drug. As the audience pointed out, her case study morbidly has a physician trying to find an article to help determine which mis-medication killed the patient instead of offering a specific warning while the patient was still alive. Speaking of post-mortems, Maastricht U is "downsizing" its Math and Computers department into, get this, Economics. Loes is getting her PhD in a year anyway. Floris is now with Amsterdam Medical Center teaching bioinformatics. I suppose Jaap van der Herik is on his way somewhere as well. Perhaps he'll be the 2nd post-doc for Passepartout? Naw, he'll probably figure something else out. I had a train ride back with UvA's David Ahn. He presented a paper on automatically extraction time from text. Sounded kind of Artequakty, but he is not one of the 15 people of have heard of Artequakt, and it's not within his circle of communities anyway. We chatted a while about extraction.