Research On Knowledge Mapping Embedding Based On Graph Network

Knowledge Graph,also known as scientific knowledge map,is a component of artificial intelligence technology.Its powerful semantic processing and interconnected organization capabilities provide the basis for intelligent information application.However,due to the explosion of Internet information,there are a large number of non-existing relationships and entities almost every day.Update messages that keep pace with the times,complementing existing relationships based on today's models,and improving user search quality and search experience are challenges in the field of knowledge mapping today.From the day the concept of knowledge map was born,a variety of groundbreaking models were proposed to solve the problem of knowledge m ap complementation.Most models map concepts to space and optimize the model based on the definition of the loss function in the model.There is also a part that takes into account local information and structure,builds the graph,and optimizes the entire graph.Based on the problems of the above methods,this paper proposes a method based on probability graph model and a computational model using the method to map variable entities and relationships in the knowledge map model into high-dimensional space.Think of each variable as a node in space and build a graph network.Using the traditional machine learning method,the whole probability map is optimized,so that each variable finds the best position in space,and then according to the distance between the entity and the relationship,it is judged whether the knowledge map triples are between There is a connection to complete the complement of the knowledge map model.For link prediction,according to the distance between the nodes,it is judged whether there is a link between the two nodes,and the network of the graph is perfected.This paper is roughly divided into three modules.1)Based on the Markov random field and the VAE model,the graph is divided into maximal groups,and then an optimization function is established to optimize the encoder and decoder parameters.2)Based on the loop confidence message propagation algorithm,the influence between each node in the graph model is regarded as a message,and the role is transmitted through the message.Through the use of the kernel function,the operations in the probability space are mapped into the embedded space,so that the embedded vector in the embedded space is directly optimized to obtain a suitable embedding position.However, according to the previous data,only the parameters between the observed variable and the hidden variable are considered.Using the score function in tranE,the distance relationship between hidden variables is established and the model is optimized.3)Based on the loop confidence message propagation algorithm,it is also the effect between the nodes passing through the message.But different from the above is to consider the role of decoding.At the end of the paper,the model is compared and tested.The comparative experiment is divided into two parts,one part is the knowledge map completion research experiment.This part is mainly compared with some popular knowledge map models,such as the Trans series model.The other part is the link prediction study,which is mainly compared with some graph embedding models,such as Deepwalk and other graph embedding models.It can be seen from the comparison test that the model has good performance in both parts.