Research on delta-radius Based Generalization of Corner Classification

Feed forward neural network based on corner classification (CC neural network, CC) can classify document instantly with cosine similarity as criteria. Usually the number of hidden neuron in a CC network equal to the number of sample documents. That is the prime shortcoming of a CC network because too much hidden neuron in a CC network will decrease its efficiency on classification instantly. TextCC is a CC network for document classification with cosine as similarity as criteria. To reduce the number of hidden neurons in TextCC, GenTextCC, the generalization of TextCC, is presented in this paper. In GenTextCC, delta- radius for each sample document is introduced and the number of hidden neurons is decreased steeply comparing to the number of TextCC"s hidden neuron. Experimental results show that GenTextCC can classify document faster than TextCC can do while the classification precision of GenTextCC is much near the classification precision of TextCC.