Procedures for the precise analysis of massive textual databases.

Abstract

Unsupervised Artificial Neural Networks have been used in the analysis of text. In general, they provide richer, deeper and more finely detailed clusters than co-occurrence models because of their ability to consider indirect connections among words. Since the number of possible indirect connections increases exponentially with increases in size of the network, this advantage should be greatly amplified in very large datasets. In this study over 4,500 world news articles about disability were gathered and analyzed using a large artificial neural network running on the Center for Computational Research (CCR)’s supercomputing cluster at the State University ofNewYork at Buffalo. Increasing the size of the artificial neural network allowed more connections between concepts to be discovered. This led to a network that was better trained and results that were more detailed and informative, showing more depth.