Geographic Information Retrieval and Digital Libraries

In this demonstration we will examine the effectiveness of Geographic Information Retrieval (GIR) methods in digital library interfaces. We will show how various types of information may benefit from explicit geographic search, and where text-based place name search may be sufficient. We will also show how implicit geographic search (or geographic browsing) can be used to dynamically generate geographic searches in geographic interfaces like Google Earth. In this demostration we will show the algorithms used for Geographic search and how these may be combined with text search. In addition we will show results from the GeoCLEF IR evaluation for text-based search. 1 Geographic Information Retrieval The goal of Geographic Information Retrieval (GIR) is to retrieve relevant information resources in response to queries with geographic constraints. GIR implies that the indexing and retrieval of objects in a digital library collection takes into account some form of georeferencing[2], and may use various forms of geographical proximity, containment, or other spatial relations in estimating or predicting relevance. Systems that provide searches using GIR methods, including geographic digital libraries, and location-aware web search engines, are based on a collection of georeferenced information resources and methods to spatially search these resources with geographic location as part of their search specifications. Information resources in digital library collections can be considered georeferenced if they are spatially indexed by one or more regions or points on the surface of the Earth, where the specific locations of these regions are encoded using spatial coordinates directly (geometrically), or indirectly by toponyms (place names). One common approach in digital libraries has been to use place names as a geographical search surrogate. However, place names have well-documented lexical and geographical problems [3]. Lexical problems include lack of uniqueness, variant names or spellings, and name changes. Geographical problems include boundaries that change over time and geographic features or areas without known place names. While geographic coordinates provide can an unambiguous and persistent method for locating geographic areas or features, they also present Fig. 1. Geographic Searching in the Incunabula Short Title Catalog (ISTC) their own set of challenges for efficient implementation. Among these challenges is the fact that the most popular interface for search systems (the simple search box), is extremely cumbersome for entering geographic searches based on coordinates. Users will seldom, if ever, know accurate coordinates for the places they are interested in. They can, however, often find them on a map. In this demonstration we will show how map-based interfaces (using Google Earth and Google Maps) can be used in conjunction with GIR search methods for retrieval of digital library information. 1.1 Probabilistic Spatial Ranking A search method that employs the “Probability Ranking Principle”, is one in which information objects are ranked and presented to the user in decreasing order of their estimated probability of relevance to the user’s information need[6]. In previous work [5, 1] we have described the development and testing of a probabilistic GIR retrieval model based on logistic regression. The form of that model used in this demonstration estimates the probability of relevance for a particular query and particular record in the database P (R | Q,D), using the equivalent “log odds” of relevance expressed logO(R | Q,D) for a set of coefficients, ci, associated with a set of S statistics, Xi, derived from the query and database, Fig. 2. Geographic Searching in the Congressional Biography Database