Meta-structure Learning Model Based On Heterogeneous Graph Transformer Network And Application Research

The review of scientific research project initiation is one of the important contents of scientific research evaluation work.It is particularly important to accurately select experts who are familiar with the research content and technical fields of the project to be reviewed for the effectiveness of the review.Based on the research of traditional expert selection methods,this thesis uses natural language processing and machine learning and other technical means to obtain semantic information,proposes a metastructure learning model based on heterogeneous graph transformer network,mines semantic information implied in the project text to be reviewed through meta-structure features,abstracts its technical field according to semantic information,and then selects appropriate expert review projects.The main completed works are as follows:(1)Analyze and study expert information,project information,and traditional expert selection methods,select information that best reflects the characteristics of the project’s technical field,and construct a heterogeneous information network through analysis of text types and their interrelationships as input for the meta-structure learning model.(2)A meta-structure learning model based on heterogeneous graph transformer network is designed,which automatically learns the meta-structure in heterogeneous information network.This model adds an automatic learning meta-graph module based on the automatic learning meta-path,which learns the semantic information implicit in project text data through meta-structure features.Then,through graph convolution networks,nodes in heterogeneous information networks are classified and clustered,and the output results are used for the classification of the projects to be reviewed.The experiment shows the effectiveness of the model in text classification tasks.(3)Developed a prototype of the expert recommendation service system for scientific research project initiation and review.Apply the meta-structure learning model based on heterogeneous graph transformer network proposed in this thesis,and design and develop relevant functions.Through the system,a batch of projects to be reviewed can be submitted,and a list of recommended review experts can be provided.This system can help project management departments find experts who are more familiar with the content and technical fields of the project to be reviewed,further improving the objectivity and impartiality of project reviews.In summary,this thesis uses the meta-structure learning model based on heterogeneous graph transformer network to obtain semantic information and classify it for application,which improves the expert recommendation method in the project review process and has certain practical application value in promoting scientific research evaluation.