|Arjen de Vries : "How to build the next 1000 search engines?"|
Fecha: jueves 29 de noviembre de 2012
Lugar de celebración: Sala de Grados A. Escuela Politécnica Superior, UAM
Years of research and development have been directed at a few search boxes only - web search now works well for general purpose requests, but what will we do about all the remaining applications that involve search? A good search engine would deal with heterogeneity in document types, in user task and experience, in target entities, and would mix evidence from the content itself and its annotations, varying from anchor text to tweet mentions. Once we drop the idea to find the "one" silver bullet retrieval model (that would perform optimally in any context, out-of-the-box), then a novel research question arises immediately: what can we do to alleviate the effort necessary to design and implement good search functionality in new application settings, i.e., to tailor the engine to the task at hand? A novel approach taken by CWI spin-off Spinque takes a two-step approach to address this challenge. A visual query environment (the "search strategy builder") allows to specify how to search. The resulting search strategy is then compiled into a search engine, backed by probabilistic relational database technology; using column-store technology to achieve the runtime efficiency needed. The talk presents the basic architecture of Spinque's platform, some of the problems addressed already, and the specific challenges ahead of us.
Arjen P. de Vries is a tenured researcher at CWI leading the Interactive Information Retrieval research group, and a full professor (0.2 fte) in the area of multimedia data management at the Technical University of Delft. De Vries studies the intersection of information retrieval and databases. He has held general and programme chair positions at SIGIR 2007, CIKM 2011, ECIR 2012 and ECIR 2014. De Vries is a member of the TREC PC (who coordinated enterprise search and entity retrieval tracks), and a steering committee member of INEX (the Initiative for the Evaluation of XML Retrieval). De Vries co-founded Spinque, a CWI spin-off that brings the Search by strategy idea to market, to provide integrated access to any type of data, customized for information specialist or end user, aiming for better and transparent search results.
Lugar de Celebración
Sala de Grados AEscuela Politécnica Superior, UAMc/ Francisco Tomás y Valiente, 11Campus de Cantoblanco28049 Madrid