Neural networks and document classification

Discusses the relevance of neural networks to the problem of classifying electronic natural language documents. We show that such documents can be represented as numeric concept vectors in a semantically meaningful way, so that AI tools such as the backpropagation learning algorithm and self-organizing maps can be used to build efficient and effective automatic document classifying systems. We show that the neural networks concerned can be taught to classify natural language text according to predefined specifications within tolerable error bounds. The convergence properties of the prototype NeuroZ described in this paper show that neural networks provide a promising platform for the automatic classification of natural language documents and that a system can be built which distinguishes in a semantically consistent way between relatively complex distinct linguistic patterns.<>