05305926

a

05305926 Leveling, Johannes Veiel, Dirk Experiments on the exclusion of metonymic location names from GIR. Peters, Carol (ed.) et al., Evaluation of multilingual and multi-modal information retrieval. 7th workshop of the cross-language evaluation forum, CLEF 2006, Alicante, Spain, September 20--22, 2006. Revised selected papers. Berlin: Springer (ISBN 978-3-540-74998-1/pbk). Lecture Notes in Computer Science 4730, 901-904 (2007). 2007 Berlin: Springer EN

doi:10.1007/978-3-540-74999-8_114

Summary: For the GeoCLEF task of the CLEF campaign 2006, we investigate identifying literal (geographic) and metonymic senses of location names (the location name refers to another, related entity) and indexing them differently. In document preprocessing, location name senses are identified with a classifier relying on shallow features only. Different senses are stored in corresponding document fields, i.e. LOC (all senses), LOCLIT (literal senses), and LOCMET (metonymic senses). The classifier was trained on manually annotated data from German CoNLL-2003 data and from a subset of the GeoCLEF newspaper corpus. The setup of our GIR (geographic information retrieval) system is a variant of our setup for GeoCLEF 2005. Results of the retrieval experiments indicate that excluding metonymic senses of location names (short: metonymic location names) improves mean average precision (MAP). Furthermore, using topic narratives decreases MAP, and query expansion with meronyms improves the performance of GIR in our experiments.