Construction And Application Of Network Domain Knowledge Graph Based On YANG Models

With the continuous development of network information technology,the Internet has become an important infrastructure highly related to the national economy and social development.Hence,the scalability and robustness requirements for the network infrastructure reach a new height.The new generation network management protocol NETCONF and YANG models used to describe network configuration and state data,aim to reduce the cost of manual operation and maintenance and make network operation and maintenance more automatic.However,due to business reality and other factors,YANG models constructed by individual vendors and standardization organizations are heterogeneous,which brings challenge to network operation and maintenance interoperability.As a conceptual modeling tool that can describe information systems at the semantic and knowledge levels,knowledge graph can achieve knowledge sharing and reuse and improve the semantic expression of information,thus providing good support for semantic transformation in heterogeneous network environment.Facing the heterogeneous network,how to build a universal network knowledge graph and use it in the network intelligent operation and maintenance has become an urgent problem to be solved.To provide a global view with unified semantics for network intelligent operation and maintenance,this thesis proposes a solution that extracts network domain knowledge and thus constructs knowledge graphs from the YANG language and models which are popularly used in the network industry.The scheme addresses the interoperability problem of heterogeneous networks and provides a unified knowledge view for network operation and maintenance management.The main works are as follows:(1)As per the YANG language specification,the approach to construct a network knowledge ontology is discussed first.Basic principles and methods for constructing the ontology are deliberated,forming an ontology structure containing 53 classes and more than 70 attributes.Then,based on the network knowledge ontology,single source knowledge graph is formed by knowledge extraction and instantiation operations including entity extraction,relationship extraction and attribute extraction for heterogeneous YANG models from different vendors and standardization organizations.In this process,it is also necessary to dynamically expand the network knowledge ontology according to the characteristics of YANG models,hence constructing a network knowledge ontology and graph that can fully express YANG's original structure and semantic information.(2)With regard to the existence of heterogeneous co-reference nodes,the node-matching problem between YANG models is transformed into the entity alignment problem between single-source knowledge graphs.Following the method of representation learning,the relational triples and attribute triples in single-source knowledge graphs can be converted into a low-dimensional vector space,with the help of embedded models and existing priori aligned entity pairs.Then,according to the semantic distance measurement between entities in the unified vector space,the entity alignment of synonymous nodes is accomplished,ending up with the knowledge graph of network domain.(3)A ptototype tool,which constructs knowledge graph from commodity YANG models,is developed,contributing a network knowledge graph comprising 347 YANG models from one vendor and two standardization organizations.This thesis discusses the application of network domain knowledge graph in network intelligent operation and maintaince,proposes an automated construction method of AIOps ontology,and accomplish the transformation from abstract network knowledge to concrete AIOps data graph through the model-driven AIOps data collection scheme.As far as we know,the approach proposed to extract network knowledge ontology from YANG language,and then construct network domain knowledge graph based on heterogeneous YANG models,is most likely to be the first in networking communities.This leads to the uniform technical solution of organizing complex and heterogeneous network configuration and state data.The methods adopted herein and the generated network knowledge graph are of theoretical and practical significance to the follow-up research and AIOps applications.