| During these years,with the rapid growth of network technology,the scale of network infrastructure and services has also increased dramatically.With the increasing complexity and heterogeneity of networks and their devices,network O&M must meet the demands of intelligence and automation brought about by the rapid network development.With the continuous expansion of the scale of converged network,the configuration management of large-scale network has ushered in great challenges.First,for traditional network topologies,operation and maintenance personnel cannot obtain more comprehensive configuration management information.Secondly,when operation and maintenance personnel need to perform end-to-end analysis of network devices on a daily basis,they need to perform complex reasoning on them.It is difficult to conduct comprehensive analysis in a short period of time,and the realtime responsiveness is insufficient.At present,although some manufacturers have studied the knowledge graph of network configuration that can support intelligent operation and maintenance,it is not strong in landing,scalability and portability.Therefore,in order to better help operation and maintenance personnel to automate operation and maintenance,it is essential to build a real-time,comprehensive,unified and highly scalable network configuration knowledge graph.Therefore,regarding the issue above,this paper first designs and implements the entity extraction method for network configuration knowledge graphs,and then designs and implements the entity extraction subsystem of network configuration knowledge graph.The research content is as follows:(1)A entity extraction method for network configuration knowledge graphs in proposed.First of all,in view of the problem that the acquisition of network configuration annotation dataset is not only time-consuming and labor-intensive,but also the network configuration dataset contains a large number of redundant samples.This paper proposes an improved active learning-based sample extraction method,which not only reduces the cost of annotation,but also reduces the cost of annotation.Resolve outliers and network configuration training set sample redundancy.Secondly,in view of the problem that different features are not considered in the network configuration entity extraction model,which will have different effects on the entity extraction results,this paper proposes an entity representation model based on an improved attention mechanism,which improves the accuracy and performance of the network configuration entity extraction model.recall rate.Finally,according to the analysis of simulation experiments,the above improved algorithm has obvious improvement in accuracy,recall and F1 value.(2)This paper designs and implements the entity extraction subsystem of network configuration knowledge graph.First of all,this paper elaborates the requirement analysis of the entity extraction subsystem of the network configuration knowledge graph from two aspects of function and performance.Secondly,this paper expounds the architecture design of the system and the workflow design of the system modules in detail.Then,this paper designs and implements a series of modules such as entity recognition,sample extraction,graph query,entity query and relation query.Finally,the results of functional test and performance test show that the entity extraction subsystem of network equipment configuration knowledge graph can efficiently and accurately perform the above operations,meeting the functional and performance requirements of the entity extraction subsystem of network equipment configuration knowledge graph. |
