The Structural Feature Of Milestone Papers And The Early Identification Of Them

In recent years,identifying papers of importance or influence in citation networks has become a hot research topic.Recent works aimed to understand how to identify a kind of special scientific papers of great significance,the milestone papers,from large-scale citation networks.To this end,previous results found that global ranking metrics that take into account the whole network structure(such as Google's PageRank)outperform local metrics such as the citation count.But these global ranking metrics are usually less in-terpretable and more computationally expensive than local metrics.Here,we analyze the local structural characteristics of citation networks.And by leveraging the recursive equa-tion that defines the PageRank algorithm,we define a family of local metrics which only utilize limited local citation information.They outperform the citation count in identifying the milestone papers,and perform nearly as well as global metrics,which demonstrates that the local variants of global metrics have the ability to maintain accuracy in detecting seminal nodes and improve the computational efficiency at the same time.Our results indicate that local metrics do not necessarily identify seminal papers worse than global ones,and they are more explainable,which could help to better understand the nature of groundbreaking research from the aspect of the network structure of citation networks.This insight can be very helpful in a particularly large network especially when the en-tire topology is hard to obtain and local network topology is easily accessible.Milestone papers have a great influence on related academic research fields and even create new re-search fields.It is of practical significance to identify such important papers at their early stage after publication.Therefore,this thesis also analyzes and compares algorithms' abil-ity to identify milestone papers at their early stage after publication.The main work of this thesis consists of following three parts:1.The basic concepts of complex networks and citation networks,the common sci-entific literature databases and the statistical and structural characteristics of citation net-works are introduced in detail.And the statistical and structural characteristics of citation networks are analyzed on two common scientific literature data sets.At the same time,we also introduce the common algorithms of assessing papers' importance in detail,as well as methods to evaluate the performance of these algorithms.2.The age bias in citation networks and its influence on the performance of ranking algorithms are introduced in detail.Several methods to suppress age bias are introduced and their effectiveness is verified on two scientific literature data sets.3.The relationship between the performance of PageRank algorithm and local struc-ture of citation network is analyzed.By leveraging the iterative process of PageRank algorithm,a series of local algorithms are constructed.The performance of this series of local algorithms in identifying milestone papers is analyzed.Among them,the 2-order local variant is compared with other global algorithms and local algorithms(Their iden-tifying performance and computation time are compared).Algorithms' ability to identify milestone papers at their early stage after publication are compared.