Neural networks aided automatic keywords selection

Abstract

Document analysis is always connected with proper selection of keywords. Many different techniques were developed. This article presents our approach to keywords generation. This approach uses Kohonen Self Organizing Maps extended with reinformement based on large library of documents, which are divided into categories (described in another articles). This article shows the differences between this method and statistical method and neural network without reinforcement. In the ending we discuss problem of algorithm performance, which can be improoved by using Genetic Algorithms, median approximation and shape-changing neural networks. © 2011 Springer-Verlag.