本体算法中相似度矩阵的学习  

Learning of Similarity Matrix for Ontology Algorithms

作　　者：吴剑章[1,2] 朱林立[3] 高炜[4] 

机构地区：[1]东南大学计算机科学与工程学院,南京210096 [2]东南大学计算机网络和信息集成教育部重点实验室,南京211189 [3]江苏理工学院信息中心,江苏常州213001 [4]云南师范大学信息学院,昆明650500

出　　处：《小型微型计算机系统》2015年第4期773-777,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60903131)资助;江苏省科技厅项目(BE2011173)资助

摘　　要：本体图中顶点之间的相似度计算是各类本体算法的本质所在.本体图中各个顶点对的相似度组成本体相似度矩阵,因此得到一个最优相似度矩阵是本体应用的实质.本文提出一种通过计算距离矩阵来得到本体相似度矩阵的方法,该方法着眼于降维过程的稀疏化和解的光滑性.从样本集得到相似顶点对集合S和不相似度顶点对集合D,由此得到三元组Γ.将Γ的信息融入到计算模型中,进而使得距离矩阵保持了原本体图中顶点间的距离结构特征.借鉴凸最小最大优化模型的光滑逼近法,得到距离矩阵计算模型的求解策略.最后,通过两个具体实验表明,本文所给的相似度矩阵计算方法对于特定应用领域中的本体相似度计算和不同本体间建立本体映射具有较高的效率.The essence of the ontology algorithm is the similarity calculation between vertices of ontology graph. The similarity of all pairs of vertices in ontology graph form the ontology similarity matrix, the essence of the ontology applications can be regarded as to get an optimal similarity matrix. This paper presents a technology to calculate the similarity matrix in terms of the distance matrix learning ,and the tricks focus on sparse in dimension reduction process and the smoothness of solution. By virtue of sample set,we ob- tain a set S consists of similarity pair of vertices and a set D consists of dissimilarity pair of vertices, and thus yield a triple set Г. By integrating Г into the calculation model, the distance matrix maintains the original structure of distance between vertices in ontology graph. In view of smooth approximation technology for convex Min-Max optimization model, the strategy for obtaining the solution of matrix calculation model is given. Finally, two experiments show that the new algorithms for specific applications with high efficiency.

关 键 词：本体 相似度 本体映射 相似度矩阵 最小最大优化 稀疏 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]