Mining Multivariate Implicit Relation In Academic Network

Multiple cooperative relationships exist widely in academic circles.Under the current environment,scientific research activities are gradually becoming complex and international.As a hot research topic in sociology and other disciplines,the mining and analysis of the relationship between tutors and students in the academic cooperation network still faces many challenges,such as the lack of general models and the difficulty in determining the multiple relationships.Most of the traditional methods of mining advisor-advisee relationship only focus on the binary relationship,while the multiple advisor-advisee relationship can play a more important role in researching academic inheritance and building academic genealogy.This paper mainly uses the MAG open-source dataset provided by Microsoft to build an academic cooperation network,and uses the Scrapy framework to crawl the AFT dataset.Due to the duplication of names in the MAG dataset,this paper specially disambiguates scholar names.The AFT data obtained by crawling is used to add label information to the academic cooperation network,and the relationship in the network is further analyzed.For the problem of lack of general model for implicit relationship mining and inefficient feature selection,this paper constructs the DAERM model by artificially designing node features and edge features,combined with the Deep Auto-Encoder,which can realize rapid mining of multi-advisor-advisee relationships in large networks.Node features represent the scholar’s own attributes,and edge features are used to describe the closeness of the cooperative relationship.For the problem that traditional methods do not fully utilize the structural features of cooperative networks,this paper proposes the GCN-Caps model.The model uses the Graph Neural Network to extract the structural features of the cooperative network,merges the three features of nodes,edges and structures,and migrates the Capsule Network to the field of multivariate relationship analysis.At the same time,the Warm Restarts technology is used to optimize the learning rate of the Capsule Network to speed up the training process,and a random deactivation process is also added to prevent the model from overfitting.The experimental results show that the relationship mining accuracy of the GCN-Caps model is high,but the model requires more computing resources and longer training time,and is suitable for relationship mining in small networks.This paper also explores the effects of network structural feature vector dimension,routing iteration times and dynamic learning rate on the performance of the Capsule Network by adjusting hyperparameters.Through the above model,this paper studies the multivariate relationship mining algorithm,which not only promotes academic network analysis,but also provides a new perspective for academic team cooperation analysis.