Identifying location-based information from the WWW, such as street addresses of emergency service facilities, has become increasingly popular. However, current Web-mining tools such as Google's crawler are designed to index webpages on the Internet instead of considering location information with a smaller granularity as an indexable object. This always leads to low recall of the search results. In order to retrieve the location-based information on the ever-expanding Internet with almost-unstructured Web data, there is a need of an effective Web-mining mechanism that is capable of extracting desired spatial data on the right webpages within the right scope. In this paper, we report our efforts towards automated location-information retrieval by developing a knowledge-based Web mining tool, CyberMiner, that adopts (1) a geospatial taxonomy to determine the starting URLs and domains for the spatial Web mining, (2) a rule-based forward and backward screening algorithm for efficient address extraction, and (3) inductive-learning-based semantic analysis to discover patterns of street addresses of interest. The retrieval of locations of all fire stations within Los Angeles County, California is used as a case study.