Access to geographic concepts in online bibliographic files: Effectiveness of current practices and the potential of a graphic interface

The focus of this research was to determine the accuracy and predictability, and hence the effectiveness, of current practices of indexing geographic concepts for retrieval from online bibliographic files within the domain of earth sciences. The methodology was based on the measurement of geographic similarity between pairs of documents. The geographic study area for each document in a test file of earth science documents was represented in at least three ways: by a map and by the text of bibliographic records from two online bibliographic files. The geographic similarity of the documents to one another was measured spatially using the maps and linguistically using the text, both indexing terminology and free text, under both Boolean and vector retrieval models and with frequency weighting of terms. Correlation analysis of the map-based geographic similarities to the text-based similarities was used to evaluate the effectiveness of geographic representation. Some records also included representation of the geographic concepts with latitude and longitude coordinates which were compared spatially to the map-based representations of the study areas and to the text-based representations. Optimal recall and precision values for three case study areas, using text and coordinates, were also derived, using the overlap of map areas to define the relevant sets. Results indicate only weak correlations between the text-based and the spatially-based geographic representations (with a range of 0.19 to 0.38), related to the imprecise nature of words in representing geographic areas and to the lack of predictability of the terminology used to describe a particular area. Recall and precision values for optimal search strategies for three case studies exhibited a great range of values (both ranged from 15% to 80%), with average values of 50% recall and 41% precision. Free text performed better than index terms in both correlation values to map-based geographic similarities and in search strategies; the advantage was based primarily on individual words in the index term phrases.