Research Of University Collaboration Set Discovery Based On Scientific Influence Analysis

Nowadays,scientific research cooperation is an indispensable part of academic research.As people continue to explore the unknown field of the world,the cross-disciplinary problem has gradually attracted people’s attention.Therefore,researchers from different research fields often work together to complete a more complex project.Scientific research collaboration usually presents the characteristics of group,that is,researchers or universities in the same group often have more research collaboration,while researchers or universities in different groups have less research collaboration.In order to effectively discover such "groups" of university collaboration,this thesis based on scientific influence analysis,combined with the random walk based network representation learning,and designed a scientific influence analysis based clustering algorithm for the problem of university collaboration set discovery(Inf2Vec-Cluster).Specifically,this thesis first uses the scientific influence analysis to learn multiple scientific influence features that incorporate different properties,and then use the Node2Vec-based network embedding to integrate these scientific influence features to learn the university embedding.Finally,this thesis uses the clustering algorithm based on K-Medoids to discovery the university collaboration set.This thesis extracted research project data from the real research management database to construct a research project network,and then conducted a comparative experiment on the extracted research project network.The experimental results show that the Inf2VecCluster algorithm,which combines scientific influence analysis and network representation learning,has a better modularity score and DBI index than other embedded algorithms and directly uses scientific influence features,which proves the Inf2Vec-Cluster algorithm can discover university collaboration set better.In addition,this thesis also conducted detailed comparison experiments and analysis of the optimization schema adopted in each step of the algorithm flow and proved the effectiveness of the Inf2Vec-Cluster algorithm from the perspective of step-by-step effects.