一种基于IDEF1x模型的层次多关系聚类算法  

A Hierarchical Multi-relational Clustering Algorithm Based on IDEF1x

作　　者：黄少滨[1] 程媛[1] 万庆生 刘国峰[1] 申林山[1] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,哈尔滨150001

出　　处：《自动化学报》2014年第8期1740-1753,共14页Acta Automatica Sinica

基　　金：国家科技支撑计划(2012BAH08B02);哈尔滨工程大学中央高校基本科研业务专项资金项目(HEUCF100603;HEUCFZ1212)资助~~

摘　　要：多关系聚类仍存在利用统计方法提取一对多联系对应的信息时会忽略数据的原始特征、不同关系表间的联系出现的回路可能导致信息重复利用等问题,且尚未见有效的解决方法.本文认为利用IDEF1x模型中不同联系的特点,可重构有助于解决上述问题的模型.因此基于IDEF1x模型构建多关系数据集中表间关联关系层次模型的框架,然后定义框架中不同种类的联系对聚类结果传递的影响,以及整合多个子节点聚类结果的方法,并以此为基础提出新的多关系聚类算法.在真实的以及人工数据集上的实验效果表明,相较于单关系聚类算法以及对比的多关系聚类算法,所提算法可获得较准确的聚类结果.There still exist some problems in multi-relational clustering, such as the primitive characters of data may be ignored when getting information reflected by one-to-many relationship through statistical methods, and information reutilization may result from the loops in the relationship between tables. We feel that a model can be rebuilt to solve these problems based on the characters of relationships in the IDEFlx model. Therefore, a hierarchical model of relationships between tables in multi-relational dataset is built based on the IDEFlx model, the effect brought by different relationships in the model as well as the integration of multiple clustering results are defined, and then a new multi-relational clustering algorithm is proposed based on these definitions. Finally, the experiments on real-world and synthetic dataset indicate the accuracy of our algorithm.

关 键 词：多关系聚类 IDEF1x模型 最短路径 结果传递 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]