GeoScience self-organizing map and concept space (poster)

The goal of this project is to develop techniques to support fuzzy, concept-based, geographic information retrieval (GIR), by proposing and prototyping a Geospatial Knowledge Representation System (GKRS) architecture which integrates multiple multi-media (textual and image) knowledge sources to support concept based GIR. Based on semantic network and neural network knowledge representations, the GKRS loosely couples different knowledge sources and adopts a spreading activation algorithm for concept-based knowledge inferencing. Both textual analysis and image processing techniques are proposed to create textual and visual geographic knowledge structures. Two of the components of GKRS, the Kohonen Self-Organizing Map and the concept space, will be demonstrated in this poster session. Kohonen-SOM: Neural network algorithms in particular, appear to be a natural starting point for organizing large amounts of information in a manner consistent with human mental models. Our research shows that the Kohonen SOM-based algorithm is able to successful create a Geo-Science based browsing tool, the Geo-MAP. It is able to loosely partition the geographic information into sub-spaces that allow users to successWIly navigate to Geo-Science abstracts of interest. For example, in the Water and Waterworks Engineering domain, the topics related to water dominate the upper-right, right hand side, and bottom portions of the map, while the terms associated with atmosphere are grouped in the upper left portion of the map. Selection of the region labeled “River” takes the user down a level where the topic “Mississippi River” can be selected and relevant abstracts found. Concept Space: To further improve the recall and precision of keyword searching within a subject category, our research group has proposed a concept space approach to Geo-Sciences information searching and retrieval. By analyzing the co-occurrence probabilities of keywords in abstracts of specific subject categories (for example, Geo-Sciences), a concept space can be created for each subject category. Such a concept space would represent the important terms and their weighted relationships in a graphical structure, akin to an associative man-made thesaurus. Our GeoSpace contains 42 Geo-Science concept spaces, allowing the user to select one or more spaces to be searched, and permits entry of up to four search terms. Our research has shown that the automatically generated GeoSciences thesaurus or concept space can be used to suggest terms within a given set of domains that are semantically related to a set of input terms. For example, a search of the concept spaces “Dams” and “Flood Control” with a user supplied term “Power Station” gives 37 related terms within these domains to assist the user in expanding or narrowing the search. DL 97 Philadelphia PA, USA Copyri&t 1997ACMO-89791~868-1/97l7..S3.50 Internet Browsing and Searching: User Evaluations of Category Map and Concept Space Techniques Hsinchun Chen, Bruce R Schatz, Andrea L. Houston, Robin R. Seweli, Tobun Dorbin Ng, Chienting Lin