Dimension reduction in text document retrieval by Hebbian neural network and nonlinear activation functions

The paper deals with utilization of neural networks for information retrieval. It is focused on reduction of text document space by Hebbian neural networks. The Hebbian neural network with Oja learning rule with linear activation function reduces term space into much lower dimension and gives good results for text document dimension reduction and retrieval. The aim of this paper is to try to increase the retrieval evaluation by F-measure that applies different nonlinear activation functions to the output layer of network. Results show better F-measure when applying other nonlinear activation functions instead of applying classical linear activation function. The results were verified on the collection of 50 documents and 100 terms, where documents were clustered into five different clusters. For each dimension the Precision, Recall and F-measure were computed and the results were depicted graphically.