Relationship Prediction And Recommendation Of Big Data In Science And Technology Based On Graph Neural Network

As a basic strategic resource formed by the primary productive force and primary driving force of a country and region,big data of science and technology is of great significance to the researchers engaged in scientific research.Therefore,studying the relationship prediction and recommendation of scientific and technological big data is of great significance to enhance the intelligent management and analysis of scientific and technological big data,promote scientific research cooperation and improve the output of scientific research results.This thesis studies the valuable knowledge in scientific and technological big data by constructing a scientific and technological knowledge graph.And using knowledge graph embedding technology,knowledge reasoning technology and recommendation algorithm,with the help of the idea of multi-task learning mechanism and graph neural network framework to predict and recommend the relationship of knowledge graphs did innovative research.The main research work of this thesis is as follows:(1)A relation prediction model based on graph neural network and multivariate relation learning(GNNRC)In order to solve the problems of high computational complexity and poor interpretability faced by existing relation prediction models,this thesis proposes GNNRC.First,let the Compl Ex model and GNN learning tasks be trained separately,and use the Compl Ex model to learn the embedding of multi-relationships in the knowledge graph.At the same time,the graph neural network is used to transmit messages on the closed subgraph of the knowledge graph and conduct knowledge reasoning.Then,the predicted scores of the two later stages are integrated.Finally,the integrated positive and negative prediction scores are linearly transformed to obtain the final prediction score to complete the relationship prediction on the knowledge graph.(2)A reinforcement recommendation model based on multi-task learning and graph convolutional neural network(MLRR)In order to solve the data sparsity problem of traditional recommendation algorithms and realize intelligent and personalized recommendation that gives different recommendation for different users,this thesis proposes MLRR.First,with the help of the idea of multi-task learning mechanism,the additional information of the items in the rating data of the recommendation module is obtained from the knowledge graph embedding model through the bridging unit.Then,the knowledge graph is transformed into a weighted graph of target users through a relation scoring function.Finally,using the embedded representation of GCN learning items and calculating the potential interest preferences of users,intelligent and personalized recommendation is well realized.(3)Recommendation system based on scientific and technological big dataThis thesis initially builds an offline practical application system—a recommendation system based on scientific and technological big data.The core functions of the system include knowledge analysis,knowledge context query,relational query and intelligent recommendation,which realized the intelligent semantic search and intelligent recommendation.The system can promote academic exchanges and scientific research cooperation,and can improve the quantity and quality of scientific research output.