摘要
首先针对在文本处理的高维矢量环境中Kohonen自组织映射神经网络的计算瓶颈问题和输入矢量空间中存在的问题进行分析,然后对随机映射(RM)和隐含语义索引(LSI)方法分别进行理论分析,提出用于文本处理的基于随机映射的加速LSI方法.试验结果表明,加速LSI方法可以在凸现原有语义联系的基础上,低代价、有效、可控地解决上述问题,极大地降低文本处理环境中Kohonen自组织神经网络的规模和计算代价.
The bottleneck problems of calculation in Kohonen self-organizing map neural network and problems of input vector spaces in the high-dimensional vector environment of text processing are analyzed first, and then on the basis of the theoretic analysis of random mapping (RM) and latent semantic indexing (LSI) method respectively, an RM-based fast latent semantic indexing method that is used in text processing is put forward. The experimental results show the fast LSI method based on original semantic links can solve the above-mentioned problems in a low-cost, efficient and controllable way. As a result, it greatly reduces the size and the computational cost of Kohonen neural network in text processing environment.
出处
《天津大学学报（自然科学与工程技术版）》
EI
CAS
CSCD
北大核心
2005年第4期372-376,共5页
Journal of Tianjin University：Science and Technology
基金
国家自然科学基金资助项目(60275020)
关键词
文本处理
隐含语义索引
自组织神经网络
随机映射
text processing
latent semantic indexing(LSI)
self-organizing neural network
random mapping(RM)
