Research On User-driven Ontology Summarization

Ontology is the upper structure of knowledge graph,which has been widely used in knowledge reasoning and query answering systems.A quick and precise construction of proper ontologies is the foundation of the above applications.Ontology reuse is one of the most common techniques applied in ontology construction.Comparing with constructing a new ontology,ontology reuse accelerates ontology construction,enhances the accuracy of ontology and reduces costs.However,the explosion of data leads to the expanding of ontology scale,which makes understanding ontology more difficult and obstructs ontology reuse.To solve the above problems,ontology summarization is proposed.Ontology summarization is the process of generating an abridged version of the original ontology,which contributes to understanding and reusing large ontologies.Ontology summarization can be divided into user-driven and task-driven.Owing to userdriven methods' extensive research and application,this paper focuses on user-driven ontology summarization.Existing user-driven ontology summarization includes two steps: ontology mapping and summarization extraction.During ontology mapping,ontology files are mapped into complex networks,thus metrics in complex network field can be used to analyze ontology structure.During summarization extraction,ontologies' structural and semantic information are used to evaluate the importance of concepts and relations to select important concepts and paths connecting them.Although some progress has been made in existing methods,there are still some problems: 1)Existing methods usually retain distinct structural information during mapping process.Yet relation restriction should be saved since it is a property of relation.2)Relation redundancies in ontologies have negative influence on ontology structure analysis and summarization extraction.Yet,existing summarization methods did not eliminate these redundancies before summarization.3)Existing summarization methods did not make full use of ontologies' semantic information,which limites summarization accuracy.This paper aims to solve the above problems.The main contributions are:(1)A relation restriction contained ontology mapping method is proposed.This paper mapped concepts and relations in ontologies as nodes and edges in complex networks.At first,the structure-irrelevant information is deleted.Then the anonymous nodes in the ontology are integrated to simplify the expression of relation restriction.Analyzing the proposed method in different ontologies quantitively and qualitatively,the result shows that the proposed method is beneficial to structural analysis and visualization.(2)A super-node based redundancy elimination method is proposed.The previous mapped network in(1)is used as input data.Firstly,the nodes equivalent to each other are considered as a super-node to transfer the ontology into a directed acyclic graph.Thus the redundancies relating to transitive relations can be eliminated by existing methods.Next,redundancies between equivalent relations and transitive relations are eliminated by vector scan algorithm.At last,we restore super-nodes and output irredundant network.Experiments on both synthetic dynamic networks and real networks indicate that the proposed algorithm can detect redundant relations precisely,with better performance and stability compared with the benchmarks.(3)An ontology summarization algorithm integrating semantic information is proposed.The irredundant network in(2)is used as input data.At first,concepts in the original ontology are transferred into vector lists according to their labels.Distances from one concept to the others are calculated according to the above vector lists and relations between concepts.After that important concepts are selected by clustering algorithm.At last,paths connecting to important concepts are selected according to the property of concepts and relations.Finally,the summarized ontology is generated.We apply the algorithm to real ontologies and evaluate the algorithm by standard important concepts and sub-ontologies,respectly.The result shows that the proposed method has higher accuracy.