Scholarly Impact Evaluation And Prediction Based On Social Network Analysis

In recent years,the study of social networks has received wide attention.As a branch of social network,scholarly network has become a hot research topic in social network analysis.The emergence and acquisition of scholarly big data provides strong support for the evaluation and prediction of scholarly impact.These studies are related to the vital interests of each scholar and institution development and provide important guidance for national research funding allocation.Scholarly impact evaluation and prediction research,including how to build the scholarly network and its structural evolution,scholarly impact inflation,scholarly impact inherent evolution mechanism and so on.Based on the evaluation method of scholarly impact,the prediction model of machine learning,and generative prediction model,this dissertation designs the corresponding solutions to the problems of scholarly network construction,scholarly impact inflation,and scholarly impact inherent evolution mechanism.The main contributions are as follows:First,design higher-order weighted paper impact evaluation algorithm,aiming at resolving the problems of the inflated scholarly impact and differentiate papers with the same score.The dissertation analyzes the relationships between citations and the actual geographical distance and leverages the relationships to weight citation networks for evaluating paper impact.By analyzing the relationship between citations and the actual geographical distance,and constructing the high-order citation network,the dissertation proposes the high-order weighted citation network.In the high-order weighted citation network,the impact of papers is quantified,and the high-order weighted evaluation method is greatly improved by experiments.Second,model scholarly paper impact based on point estimation,aiming at the inherent evolution mechanism of scholarly paper,the dissertation proposes a prediction model based on point estimation.In citation networks,the relationships between early citer's impact and citations are analyzed.By using the positive relationships between early citers' impact and citations,the change trend of long-term citations,and current citations enhancing mechanism,the dissertation models scholarly paper impact to improve the prediction results.Third,weighted citation institution impact evaluation method.Compare to paper impact,institution impact suffers more from manipulation which leads to impact inflation.This dissertation proposes an evaluation method with weighted citation,aiming at resolving the inflated scholarly impact.By analyzing the explicit and implicit Conflict of Interest relationships between scholars and weighting citations,this dissertation demonstrates an objective evaluation method to improve the feasibility of evaluation.Fourth,implicit multi-feature institution impact model.Aiming at the inherent evolution mechanism of institution impact.This dissertation proposes a dynamic temporal prediction model of institution impact based on implicit multi-feature learning.In citation networks,the factors driving citations change are analyzed.Scholar' impact,geographic position information of institutions and national GDP are used to model the predictive model.In addition,the importance scores are given for all features.The model improves the prediction accuracy.