GIRSA-WP at GikiCLEF: Integration of Structured Information and Decomposition of Questions.
Sven Hartrumpf, Johannes Leveling · 2009
This paper describes the current GIRSA-WP system and the experiments performed for GikiCLEF 2009. GIRSA-WP (GIRSA for Wikipedia) is a fully-automatic, hybrid system combining methods from question answering (QA) and geographic information retrieval (GIR). It merges results from InSicht, a deep (text-semantic) open-domain QA system, and GIRSA, a system for textual GIR. For the second participation (the first participation was for the pilot task GikiP 2008), the GIR methods were adjusted by switching from a sentence-based retrieval to an abstract-based retrieval. Furthermore, geographic names and location indicators in Wikipedia articles were annotated before indexing. The QA methods were extended by allowing more general recursion with question decomposition. In this way, complex questions, which are frequent in GikiCLEF, can be answered by first answering several depending questions and exploiting their answers. Two new resources of structured information from Wikipedia were integrated, namely the categories assigned to articles and the infobox file from DBpedia, which is an automatic information extraction approach for Wikipedia data. Both resources were exploited by