Abstract: In order to effectively reduces the cost of time and space of text processing, with reference to the theory that only a part of signal from brain cells can reach pallium, and the theory that axon signal strength is reduced with distance increment from main body of neural cells, signal transmission theory-based dimension reduction algorithm of neural network is presented. This algorithm optimizes the structure and training algorithm of neural network and points out that there exists a great number of component approaching 0 in LTM between neurons in text space, that is there exists many unnecessary connection between neurons. So the basis of text dimension reduction is formed based on this discovery. The equivalent to and even higher classification accuracy is possessed in dimension reduction space than that of original space at the lower cost of time and space by the above algorithm.

Key words: neural network, text, dimension reduction, signal transmission

摘要： 为有效降低文本处理的时间与空间代价，根据“只有部分脑细胞发出的信号能到达大脑皮层”和“突触信号强度随着与神经细胞主体距离的加大而减弱”的理论，提出基于信号传递理论的神经网络降维算法。通过神经网络结构与训练算法的改变，在文本处理环境中神经元间LTM向量中有大量逼近0的分量，即存在很多不必要的神经元连接，以此作为文本降维的基础。实验结果证明，降维后的文本数据库以较低的时间代价具备与降维前相当甚至更高的分类准确率。

关键词: 神经网络, 文本, 降维, 信号传递

CLC Number: 

• TP183