基于超图结构的知识相似度计算模型研究

A Model for Knowledge Similarity Metrics Based on Hypergraph Structure

知识表示是知识管理问题的基础,关于知识表示的方法可以划分为基于特征和基于结构两类,其中基于结构的方法支持知识对象内部关联结构的描述,近年来受到广泛的重视,但已有研究大多以经典图论作为形式化基础,其局限性在于对多元关系知识表示上存在不足,为此本文引入了经典图的泛化——超图,并对其基本概念进行扩展,定义了超图结构以进行多元关系知识的表示,在对对应、相邻度等相关概念进行定义的基础上提出了一种基于超图结构的知识相似度计算模型,其计算步骤包括知识对象可比分析、对应求解以及相似度计算,最后通过两组实验验证了模型的有效性。

Knowledge representation is the foundation of knowledge management issues,and the methods of knowledge representation could be classified into two categories as feature-based and graph-structured.Graph- structured representations are capable of expressing the inherent relations between knowledge objects,and have attached great attentions recently.However, most of researches are based on classic graph theory,which is insufficient for the representation of knowledge where N- ary associations exist.In this paper we introduce and extend hypergraph,the generalization of classical graph,and propose the definition of hypergraph structure to represent knowledge.Based on some related definitions such as corresbondence and degree of adjacency,a model of similarity measure for hypergraph structure is proposed,the computational procedure of which is consists of comparability analysis,correspondence solving,and similarity degree computing.Finally the usefulness of our method is verified by two groups of experiments.