The Self-Organising Map has been frequently employed to organise collections of digital documents, especially textual documents.
SOMs can be employed to analyse the content and relations between the
documents in a collection, providing an intuitive access to large collections.
In this paper, we apply this approach to analysing documents from the
Internet platform WikiLeaks. This document collection is interesting for such an analysis for several aspects. For one, the documents contained cover a rather large time-span, thus there should also be an quite divergence in the topics discussed. Further, the documents stem from a magnitude of different sources, thus different styles should be expected.
Moreover, the documents have very interesting, previously unpublished
content. Finally, while the WikiLeaks website provides a way to browse
all documents published by certain meta-data categories such as creation year and origin of the cable, there is no way to access the documents by their content. Thus, the SOM o ers a valuable alternative mean to provide access to the content of the collection by their content.
For analysing the document collection, we employ the Java SOMToolbox
framework, which provides the user with a wealth of analysis and interaction methods, such as different visualisations, zooming and panning, and automatic labelling on different levels of granularity, to help the user in quickly getting an overview of and navigating in the collection.