Academic Cooperation Analysis Based On Graph Convolution Network

Scholarly data has attached attentions for years.As the results of studying accu-mulating,the amount of scholarly data grows rapidly.In this circumstance,scholarly data analysis is meaningful.For the academic world,this kind of analysis could help scholars know the trending of different disciplines;to each scholars,they could use it to evaluate the importance of his/her works,and choose coauthors.Therefore,there are intensive study in scholarly data.At present,some studies about scholarly coop-eration network has been studied,such as prediction of citations,recommendation of citations,recommendation of coauthors.However,few of them using scholarly net-work embedding method,and few of them using information in scholarly cooperation.In fact,scholarly cooperation is important in scholarly network analysis,especially for the growing of beginners in scholar world.Academic cooperation analysis is an impor-tant part in scholarly data analysis,which need a more comprehensive scholarly network embedding method.After studying the researches in network embedding and scholarly network analy-sis,we find that there are few methods about scholarly network embedding,especially methods using cooperation information.In line with correlation analysis in the scholarly cooperation network,it is obvious that the authors of one paper tend to contribute this paper differently.The contribution credit is related with the similarity of study fields of author and the topic of this paper,and the influence of authors.The structure of citation network also contributes to paper embeddings.Graph convolutional network(GCN),as an graph learning model,could be used to learn this structure information.Therefore,in this thesis we analysis coefficients between the basic features in scholarly network and the scholarly output,and build a scholarly network embedding method using attention mechanism and GCN to capture different contributions among authors and structure features respectively.To fully utilize the cooperation information in scholarly network,and get a comprehensive embedding of network,here we study the scholarly coopera-tion analysis and embedding methods.The main contributions are as follow:1.Analysed the relation factors of influence of authors and papers in scholarly network,combining structure information and attributes of authors or papers;2.Pro-posed an Author Contributed Representation for Scholarly Network(ACR-SN)model,this model using attention method to combing authors embeddings and using graph con-volutional network to utilize the structure of citation network,to form the embedding of papers;3.Validated the effectiveness on scholarly data sets.The results shows that ACR-SN performed well on scholarly network data,gained higher accuracy compared with other embedding methods.