View in EDS

Geographic Information Retrieval (GIR): Searching Where and What.

Saved in:

Bibliographic Details
Title:	Geographic Information Retrieval (GIR): Searching Where and What.
Authors:	Larson, Ray R.¹ ray@sherlock.berkeley.edu, Frontiera, Patricia² pattyf@regis.berkeley.edu
Source:	SIGIR Forum. 2004 Special Edition, p600-600. 1p.
Subjects:	Geographic information systems, Information retrieval, Information storage & retrieval systems, QUERY (Information retrieval system), Algorithms, Search engines, Electronic information resource searching
Abstract:	This article reports on a demonstration of the Geographic Information Retrieval (GIR). The demonstration will show how explicit geospatial metadata and how inferred geographic information from texts can be exploited to provide effective and accurate ranked retrieval of geospatial information and relevant text documents using retrieval algorithms based on logistic regression with weighting coefficients estimated from a set of training data. A system that combines conventional probabilistic algorithms for text retrieval with algorithms for estimating probability of relevance for geographic spaces will be presented. Also to be demonstrated is the algorithm for GIR ranking that estimates probability of relevance based on a weighted set of parameters where the weights were derived using logistic regression from samples of a test collection. The demonstration will show: How graphical geospatial query specifications can be used to obtain sets of geospatial data ranked by probability of relevance; how different representations of the underlying data extents, including minimum bounding rectangles and convex hulls, compare to complex polygon representations in retrieval; how other characteristics, such as contextual geographic information, can be combined with knowledge of the query and candidate regions to improve retrieval effectiveness and how online gazetteer can be used to apply geographic retrieval to texts. The demonstrations will show real-time live searchers and geographic displays to illustrate the algorithms and methods described.
Database:	Engineering Source

Description
Abstract:	This article reports on a demonstration of the Geographic Information Retrieval (GIR). The demonstration will show how explicit geospatial metadata and how inferred geographic information from texts can be exploited to provide effective and accurate ranked retrieval of geospatial information and relevant text documents using retrieval algorithms based on logistic regression with weighting coefficients estimated from a set of training data. A system that combines conventional probabilistic algorithms for text retrieval with algorithms for estimating probability of relevance for geographic spaces will be presented. Also to be demonstrated is the algorithm for GIR ranking that estimates probability of relevance based on a weighted set of parameters where the weights were derived using logistic regression from samples of a test collection. The demonstration will show: How graphical geospatial query specifications can be used to obtain sets of geospatial data ranked by probability of relevance; how different representations of the underlying data extents, including minimum bounding rectangles and convex hulls, compare to complex polygon representations in retrieval; how other characteristics, such as contextual geographic information, can be combined with knowledge of the query and candidate regions to improve retrieval effectiveness and how online gazetteer can be used to apply geographic retrieval to texts. The demonstrations will show real-time live searchers and geographic displays to illustrate the algorithms and methods described.
ISSN:	01635840