Citation-based Similarity Measure Over Journal Citation Network

With the development of network technology and communication technology,complex networks emerge in real word.The issues of how to describe complex network characteristics draw extensive research interests.There are undirected networks and directed networks.People have carried through comprehensive researches in topology structure and model of undirected network,while the research on directed network is still at an early stage.In this thesis,we attempt to effectively compute similarities between entities within the context of journal citation networks,which is fundamental and is used for many different tasks,such as clustering,nearest neighbor classification,anomaly detection,and similarity query.We point out that the relatedness measure,RF,introduced by Pudovkin and Garfield ignores the impact of journal quality on cites,and underestimate the similarity between target journal and journals with low quality,and introduce the improved algorithm,RF'.There has been little written on how to quantitatively evaluate the accuracy of relatedness measures.The author proposes a new framework for assessing the performance of relatedness measures.RF is proved to be an effective recognition degree measure,which is size-dependent.We apply normalized RF to PageRank as transition probability and introduce measure RD Journal Rank(RDJR)to detect prestigious journals.The analysis of journals in Statistics&Probability category indicates that RDJR outperform five-year impact factor.