多关系图的多类标分类关键技术研究

项目名称： 多关系图的多类标分类关键技术研究

项目编号： No.61272538

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 叶允明

作者单位： 哈尔滨工业大学

项目金额： 80万元

中文摘要： 基于各种现实和网络虚拟空间，人与人之间形成了复杂的多关系社会网络。如何判定这个复杂网络中每个人的兴趣集合是许多应用的基础问题，可以形式化的映射为多关系图的多类标分类问题。其研究难点是：多关系图中节点、关系和类标集具有复杂的依赖关系和语义关联性,如何有效挖掘并利用隐藏在这些要素之间的关联信息是提高分类性能的关键。围绕着这个核心问题，课题将提出基于张量的多关系图表示模型，并在此基础上重点研究以下内容：(1)基于高阶马尔科夫链的类标传递要素计算方法，用于解决类标传递性计算问题；(2) 基于内容属性与多关系图拓扑结构信息融合的学习方法，用于解决类标之间依赖性的学习问题；(3)静态和增量式多类标分类方法，用于解决大规模多关系图的动态分类问题。课题的创新在于：基于张量的多关系图多类标分类模型；基于高阶马尔可夫链的类标传递计算方法；基于内容属性与多关系图拓扑结构的类标间依赖性学习方法。

中文关键词： 多关系图；多类标分类；高阶马尔科夫链；张量；

英文摘要： With various physical and virtual spaces, the relation between people becomes a complex multirelational social network. How to identify the interests for each person in this complex network is a fundamental problem in many real applications, which can be formaluated as a problem of multi-label classification of multirelational graph. The big challenge of this problem lies in the complex dependency and correlation among vertexs, edges and labels, and the classification performance will highly depend on the effectiveness of exploiting these correlations and dependencies. This project will first propose a tensor based representation model for multirelational graph, and then explore the following research issues: (1) label set propagation methods via high-order Markov chain model; (2)learning various dependencies of label set by exploiting both content information and topology information among the network; (3) both static and incremental multi-label classification method for large-scale multirelational graph. The main innovations of this proposal are as follows: a tensor based multi-label classification model; a high-order Markov chain based method for label set propagation problem; a label dependency learning approach through the combination of content and topology information.

英文关键词： multirelational graph；multi-label classification；high-order Markov chain；tensor；