A new approach to build a geographical taxonomy of adjacency automatically using the latent semantic indexing method

In this paper, we introduce an approach for constructing a geographical taxonomy of adjacency for a country, to be used in reformulating spatial queries. The proposed approach uses the best-ranked documents retrieved by the search engine while submitting the spatial entity composed of a spatial relationship and a noun of a city A. Then, apply to it the Latent Semantic Indexing method to found the nearest cities B_i to A, and proceed to a step of validation of each link by verifying if A is also found in the results of the cities B_i. In our experiments, we constructed a geographical taxonomy of adjacency for Morocco. We varied the spatial relationship used in the step of documents retrieving to compare the results of the different spatial relationships, and we used google web services as a search engine to compare the results returned in every case. Then we used the constructed taxonomy in geographical query reformulation. We have used the Un-interpolated Average Precision (UAP) to compare the returned documents before and after reformulation. According to our results, we note that reformulating geographical queries based on our built taxonomy improves widely the precision of the queries.