Exploring term selection for geographic blind feedback

Abstract

Standard blind feedback techniques have been applied to geographic information retrieval (GIR) with limited success. To facilitate more geographically oriented blind feedback, this paper explores the use of meronymy (part-whole) relations for term selection, based on a meronymy predicate for selecting feedback terms.Different methods are evaluated on the German GeoCLEF 2006data, measuring standard evaluation measures such as mean average precision (MAP) or precision at five documents. To achieve a trade-off between coverage and ambiguity, a new geographic knowledge base is created, containing important place names only, which reduces ambiguity. Results indicate that blind feedback improves performance for some topics, but does not considerably change MAP in the setup. However, a higher precision at five documents is achieved if more geographically oriented term selection criteria are applied.