基于复合结构的知识库分类体系匹配方法

发布时间：2018-05-05 01:14

本文选题：知识库 + 分类体系匹配　；参考：《计算机研究与发展》2017年01期

【摘要】：近年来,分类体系匹配由于其在知识库构建和融合等方面的广泛应用,已成为国内外工业界和学术界的研究热点.然而,随着网络大数据的不断发展,分类体系变得越来越庞大和复杂,构造一种通用有效的分类体系匹配器以适应大规模、异构分类体系匹配的扩展性仍然面临很大的挑战.为此,提出了一种基于复合结构的分类体系匹配方法 BiMWM,该方法利用分类体系中分类的复合结构信息:微观结构和宏观结构,将分类体系匹配问题转化为二部图上的优化问题进行求解.首先,创建赋权的二部图建模分类体系之间候选的匹配类对关系;然后,通过计算二部图上的最大权匹配剪枝选择最优的分类体系的匹配类对.BiMWM方法可以在多项式时间内为2个分类体系产生最优匹配.实验结果表明:与当前先进的基准方法相比,该方法能够有效提升大规模、异构分类体系匹配的性能.
[Abstract]:In recent years, classification system matching has become a research hotspot in industry and academia because of its wide application in knowledge base construction and fusion. However, with the continuous development of network big data, the classification system becomes more and more large and complex. To construct a general and effective classification system matching machine to adapt to large scale, the scalability of heterogeneous classification system matching is still facing a great challenge. In this paper, BiMWM, a matching method of classification system based on compound structure, is proposed. The method uses the information of composite structure in classification system: microstructure and macroscopic structure. The matching problem of classification system is transformed into an optimization problem on bipartite graph. First, create candidate matching class pairs between weighted bipartite graph modeling classification systems; then, By calculating the maximum weight matching pruning on bipartite graphs and selecting the matching class pairs of the optimal classification system, the BiMWM method can generate the optimal matching for the two classification systems in polynomial time. The experimental results show that this method can effectively improve the performance of large-scale heterogeneous classification system matching compared with the current advanced benchmark method.
【作者单位】：中国科学院信息工程研究所;中国科学院网络数据科学与技术重点实验室(中国科学院计算技术研究所);
【基金】：国家“九七三”重点基础研究发展计划基金项目(2012CB316303,2013CB329602) “核高基”国家科技重大专项(2013ZX01039-002-001-001) 国家自然科学基金项目(61303056,61402464,61402442,61572469,61502478) 北京市自然科学基金项目(4154086)~~
【分类号】：TP391.1

【相似文献】