Information retrieval in law using a neural network integrated with hypertext

The authors present a neural network that is designed to work with hypertext systems in applications involving the manipulation of large quantities of text. Use is made of a network in which text units are linked to significant words found in them, the weight of links being determined through an automatic text analysis based on a normalized word frequency measure. There is no intermediate layer, making it possible to construct the initial state of the network rapidly, and to readily accommodate new documents as they arrive over a period of time: a peculiar requirement of the application. Emphasis is paid to the method in which the user can provide feedback based on the value that the user attaches to the documents retrieved in response to a query