The map metaphor for displaying the contents of a document library in a two-dimensional display has gained considerable interest [1,4,5,6,7]. Maps are used to visualize the similarity between documents in terms of distance within the two-dimensional map display. Hence, similar documents may be found in neighboring parts of the map display.
The underlying assumption in all of the above mentioned papers is that the data items, i.e. the documents, are available locally for self-organizing map training. With ever growing information repositories, however, this assumption is questionable. One should rather expect that several repositories exist at different sites and a way for unified access should be provided for the users of such distributed digital libraries. Moreover, most of the reports on the application of self-organizing maps for the organization of document libraries propose the usage of a single map for document space visualization. It is only natural that with increasing size of the document library the self-organizing maps for representing this library grow larger, see e.g. [4]. This, however, does not necessarily improve the organization of the library. In particular, it is rather difficult for the user to find her orientation in large maps. To use an example from geography, it is much easier to find the way, say, from Stockholm to Skövde by using the map of Sweden than by using the map of Europe or, even worse, that of the World. In much the same way it is easier to locate a particular document using a map of a smaller portion of the library while relying on higher aggregated maps for overall orientation.
In this paper we suggest the utilization of individual self-organizing maps for organizing the different parts of a distributed digital library. Multiple maps may then be integrated to produce an aggregated visualization of the complete document space. We demonstrate the effects of such an approach based on a sample library of text documents.
The remainder of this paper is organized as follows. In Section 2 we give a brief review of self-organizing maps and describe their integration. Section 3 is dedicated to a description of our experimental document library. Experimental results of our approach to digital library organization are given in Section 4. Finally, our conclusions are contained in Section 5.