Hypertext-like structures through a SOM network

In this paper we describe a system whose main aim is supporting a hypertext author to classify and organize a large amount of documents; the system allows the author to have access to the documents with hypertext features, providing some access points and suggesting, for each document, the related ones. The system is an interesting application of the Self Organizing Map network, a neural network widely used to organize multidimensional data; specifically, it is based on two SOM networks, the first one is aimed at organizing collections of documents in `information maps' that display the relations between the content of the documents; the second one identifies access points and splits the maps into meaningful areas. Finally the author can edit both the list of access points and the map through a Web page editor, thus moving the misclassified documents in the right area.