Cyberspace geography visualization
The World-Wide Web can be depicted as a graph. Each resource is a vertex and the links are the edges. The distances between pairs of resources is then defined as the shortest path in the graph between them, leading to the creation of a metric. With the ability provided to measure the distances among resources, it becomes possible to represent each resource as a point in a high dimensional space where their relative distances are preserved.
It is clear that a high dimensional space cannot be visualized and thus its dimensionality has to be reduced. To perform this task, the self-organizing maps algorithm is used because it preserves the topological relationships of the original space, conjointly lowering the dimensionality. This creates the ability to map any resources onto a lower dimensional space, while maintaining their order of proximity.
During this non-linear dimensionality reduction, the distances among resources are lost. Since it is primordial that the distances can be evaluated, the unified matrix method is used. By geometrically approximating the vector distribution in the neurons of the self-organizing maps, this method provides a means to analyse the landscape of the mapping of cyberspace.
To permit exploratory analysis of the self-organizing map, the mapping is made onto a two-dimensional visualization media. Note, however, that reduction is also possible, using the proposed method, to a space having an arbitrary dimension. This approach enables the visual display of virtual locations of resources on a landscape, in a fashion similar to geographical maps.
A prototype performing the above task has been developed. Using real information about resources available in the World-Wide Web and their connective structure, various maps have been constructed. Given that the development is in the prototyping stage, it has been possible only to construct maps exhibiting limited numbers of resources. The visualization, comprising some interaction possibilities, is directly made available on the World-Wide Web using forms and sensitive maps, which enable direct retrieval of the resources represented on the maps.
Despite some scalability problems with the current implementation, new developments will soon handle the limitation in information gathering. An implementation model for the construction of the maps on a parallel computer has been proposed. Certainly further improvements are therefore feasible.
The results are encouraging. No major flaw has been detected in the proposed model, and the first users are enthusiasts. It is thus advocated that further research should be done in this direction.
The above mentioned results, including the documentation, are available at
<URL:http://heiwww.unige.ch/girardin/cgv/>Thousands of accesses to these maps, which show what we would like to call the geography of cyberspace, have already been reported....