Research On Network Representation Learning Method For Knowledge Graphs

With the development of Internet technology and digital economy,the network representation learning method based on deep learning technology to process network structure data has attracted the attention of academia and industry.It aims to represent nodes in the network as low-dimensional dense real-value vectors and effectively preserve network structure and other valuable information.Most of the existing network representation learning methods rarely consider the attribute information of the network when learning the node representation,and the network attribute information often contains very valuable information.The knowledge graph is a data model based on graph structure,which can naturally describe the network structure data that is widely connected between entities in the real world,and can perfectly express the attributed network in the real world.These network attribute information can not only reflect the network generation process,but also reveal the deep potential information of the network.Therefore,the study of such attributed networks has important value.This paper proposes a novel structural role enhanced attributed network representation learning model framework Rol EANE for knowledge graphs,which can effectively preserve network topological structure and attribute information at the same time.The model framework consists of two main parts: one is the network structural role proximity enhanced deep autoencoder model that can be used to capture highly nonlinear network topological structure and attribute information,and can also retain the global information of the network structure;the other part is the Skip-Gram model of neighbor strategy optimization,which can integrate the network topological structure and attribute information seamlessly and preserve the local information of the network structure.By analyzing the nature of structural role proximity,a structural role proximity optimization strategy based on information entropy and directed graph is proposed.From the perspective of the structure and implementation principle of the Rol EANE model framework,two structural role enhancement strategies for deep autoencoder,including target enhancement strategy and error enhancement strategy,are designed.The representation learning output mode of the Rol EANE model framework is analyzed,and two representation learning output strategies are designed,which are concatenate output strategy and integrated output strategy.On the four real datasets,the proposed method and its extended version are compared with several existing state-of-the-art network representation learning methods to perform three different experimental tasks: node classification,link prediction and network visualization.The experimental results show that the Rol EANE model framework has significant advantages.At the same time,the results of each experimental task show that the error enhancement strategy improvement effect in the structural role enhancement strategy is better than the target enhancement strategy,and it show that the improvement effect of the representation learning output strategy has a greater relationship with the sparsity of the network dataset.Finally,the parametric analysis proves that the Rol EANE model framework has good stability.