Evaluation And Prediction Of Scholars' Scientific Impact In Heterogeneous Academic Networks

In recent years,the rapid development of network and data analysis technologies has led to the exponential growth of scholarly big data,and it has become a new research hotspot.As an important branch of scholarly big data research,the study of scientific impact not only pro-vides powerful basis for cultivating scientific talents,but also is very crucial for evaluating the scientific progresses of nations.Therefore,the research on scientific impact is very significant.The talent issue not only relates to the regulation and control of education or research resources,but also plays a decisive role in the development of comprehensive national strength.As an important part of talents,this paper conducts an in-depth research on scholars' scientific impact.However,there are some deficiencies in the existing research on scholars' impact,such as ne-glecting the importance of scholars' cooperative relationships,the evolution and heterogeneity of academic networks,and the changing trends of the influence of different scholars.Inspired by the above-mentioned key issues,this paper utilizes scholarly big data,computational social science,and complex network theory and methods to evaluate and predict the scientific impact of scholars,excavate factors that affect the success of scholars,and predict their true ages.The main research contents and contributions of this thesis are as follows:1.Evaluation of scholars' impact.Considering academic cooperation has an important in-fluence on scholars' impact,a novel method of scientific impact evaluation is proposed based on the scholars' positions in cooperation networks is proposed.On the one hand,it applies the structural holes theory and information entropy theory to propose and define new features that quantify the influence of scholars' network location;on the other hand,it constructs a variety of heterogeneous academic networks and considers the interactions between entities in the aca-demic network to evaluate scholars' impact.It can improve the rationality of scientific impact evaluation issues.2.Prediction of scholars' impact.Aiming at the deficiency of existing forecasting methods in capturing the evolution of scholars' impact,this paper proposes a novel method for predicting the impact of scholars.On the one hand,the proposed method first divides scholars into different types according to their own characteristics;on the other hand,it considers the time evolvement characteristic of academic networks,and builds specific academic networks for different types of scholars to predict their impact.The accuracy of the prediction results can be improved.3.Discovering factors that influence the academic success.In view of the deficiencies in the current research work in mining the decisive factors affecting the academic success,this pa-per classifies the influencing factors into five types,which are article-centered,author-centered,journal-centered,institution-centered,and time factors,and applies machine learning algorithms to solve this problem.Through experiments on the real dataset,it is found that the average num-ber of citations by scholars,the number and diversity of collaborators,and academic ages are highly relevant to scholars' future success.In addition,the experiments also verify the fact that there exist the phenomenon of "groups of people" within the same institution.4.Prediction of scholars' true ages.Aiming at the important effect of scholars' ages on their own scientific impact,and the difficulty in obtaining the relevant data,a scholar's age prediction method based on representation learning algorithm is proposed.This method first uses the scholar's articles' information to extract and analyze factors that influence the age of scholar.Secondly,it combines the above factors with machine learning algorithms to predict scholars'ages,and analyze their importance.This method can protect the privacy of scholars to some extent while improving the accuracy of prediction.