Comparison of two neural networks approaches to Boolean matrix factorization

In this paper we compare two new neural networks methods, aimed at solving the problem of optimal binary matrix Boolean factorization or Boolean factor analysis. Neural network based Boolean factor analysis is a suitable method for a very large binary data sets mining including Web. Two types of neural networks based Boolean factor analyzers are analyzed. One based on feed forward neural network and second based on Hopfield-like recurrent neural network. We show that both methods give good results when processed data have a simple structure. But as the complexity of data structure grows, method based on feed forward neural network loses the ability to solve the Boolean factor analysis. In the method, based on the Hopfield like recurrent neural network, this effect is not observed.