节点文献
CC4神经网络的分类性能理论分析
THEORETICAL ANALYSIS OF CLASSIFICATION PERFORMANCE OF CC4 NEURAL NETWORKS
【摘要】 本文在提出规范、规范满等概念的基础上,对CC4神经网络分类计算的倾向性进行了理论分析。并针对文本分类,提出了基于神经网络的增量式索引建立方法,将以词频为基础表示的高维文本信息映射到低维数据空间。为了使CC4神经网络应用到基于文本信息空间索引的分类技术中,将空间索引变换为CCA神经网络可以接受的二值向量,使得CC4神经网络以空间索引为基础,进行文档分类。最后给出了相应的实验结果。
【Abstract】 On the basis of the concept of Standardization and Standard Fullness, the computational bias of CC4 networks is analyzed theoretically. For text classification, a neural network based incremental indexing method is proposed to map the high dimensional text information represented as term frequency vectors to lower dimensional data points. To make CC4 neural networks be able to be applied in document classification with their spatial indexing, indexes are transformed into binary sequence as the input of CC4 neural networks so that CC4 can perform document classification. The sensitivities of the classification precision to the generalization radius of CC4 networks is also analyzed theoretically. Finally, experimental results are given to demonstrate the correctness of our theoretical analysis.
【Key words】 Classification; CC4 Neural Network; Computational Bias of Network; Input Extension;
- 【文献出处】 模式识别与人工智能 ,Pattern Recognition and Artificial Intelligence , 编辑部邮箱 ,2003年01期
- 【分类号】TP183
- 【被引频次】2
- 【下载频次】58