Automatic Geographical Hypertext 'Multi-scaled Links' Generation

摘要

In this work, we have presented a method and described the process for automatically link text and graphics (especially maps) in a geographic context with a "multi-scaling" linking artifact. Using standard Information Retrieval techniques added with a "psycho-cognitive" like aspect, and local natural language processing, we have shown that the method is computational effective. The centerpiece of the implementation is the text processing tool, we have described aspects to model the whole processing of geographical maps. We do not claim to have a powerful solution for finding all "implicit" lonks. Information retieval techniques are perfectible the same remarks can be applied for local natural language processing. What our method point out is that a the double processing text and map is able to handle some fine distinctions in meaning improving the quality of the results. Our perspectives are in the short-term to improve the Global Link generation taking into account more precisely the distance between graphs and paragraphs, as for example : is the graph above a paragraph ?, under a paragraph ?, ... But the most interesting work in progress is the development of a common formal representation that allowed to described both analysis and interpretations of comments and maps in a same framework.