语义核SVM结合改进EMD跨越语义鸿沟  

Crossing semantic gap by semantic kernel SVM combined with improved EMD

作　　者：徐炎 曹春萍[1] XU Yan;CAO Chunping(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院

出　　处：《轻工学报》2019年第3期77-83,共7页Journal of Light Industry

基　　金：国家自然科学基金资助项目(61472256)

摘　　要：针对跨越语义鸿沟方法中未考虑文本间语义相关性和样本数量增加时计算量过大的问题,提出了一种语义核SVM结合改进EMD跨越语义鸿沟方法.该方法首先考虑到文本特征间的语义关系,提取与图像共生的文本关键词,结合HowNet通用本体库和内部统计特征构造语义核函数,然后将语义核函数嵌入SVM进行关键词分类,得到最佳候选关键词,从而解决文本间语义相关性问题;再通过最佳减小矩阵对EMD算法进行改进,从而减小计算量.对比实验结果表明,该方法充分利用了与图像共生的文本特征间的语义关系,标注准确率明显高于其他3种方法,且标注时间缩短为其他方法的1/5左右.Aiming at the problem that the semantics relation among texts is not considered and the amount of computation is too large while samples increases in crossing semantic gap methods,a method of crossing semantic gap was proposed based on semantic kernel SVM combined with improved EMD. Firstly, to solve the semantic relation problem among texts, the method constructed the semantic kernel function based on taking semantic relations of text features into consideration, extracting text features coexisting with images and combining HowNet common ontology repository.Then the semantic kernel function was embedded into the SVM to classify keywords for obtaining best candidate words. Secondly, the algorithm improved EMD with best decreasing matrix to cut down the amount of computation. The experiment result showed that the method proposed takes full consideration of semantic relation in the texts related,the annotation accuracy rate was obviously higher than the other 3 methods and the annotation time was cut down to 1/5 of before.

关 键 词：语义鸿沟 语义核 支持向量机 改进EMD 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]