基于深度自编码器的学术期刊影响力排序与分区方法

目的/意义]期刊影响力排序和分区是评价期刊影响力的重要指标。然而,现有学术期刊分区方法只使用单个或少数期刊表征因素来进行排序,从而不能反映出合理的期刊分区。利用深度自编码器提出一种新的多因素综合体系学术期刊排序方法并实施期刊分区。方法/过程]首先,利用相关系数矩阵和方差膨胀因子挑选了若干个具有高独立性的关键期刊因素;其次,使用渐进式深度自编码器构架设计策略,分析了在采用不同构架时的期刊分布规律,并将第一个隐元作为排序度量来实施期刊排序;最后,实现了两种深度自编码器分区方法,分别对比了采用平均期刊数目划分和非平均期刊数目划分的分区方法。结果/结论]选择“图书馆学;情报学”“法学”和“体育学”三类期刊为实证研究。结果表明,该方法不仅能够实现期刊全局和局部关系的多层次分析,而且能够以非线性方式将多个期刊表征因素融合为单个排序度量分值。

Impact Ranking and Partitioning Approaches for Academic Journals Based on Deep Auto-Encoder

Purpose/Significance]Ranking and partitioning of academic journals are two key factors to evaluate their impacts.Existing approaches rank academic journals by one or two journal properties and group them into different partitions.This paper proposes a novel ranking approach of academic journals based on Deep Auto-Encoder(DAE)which integrates multiple journal properties,and performs journal partitioning by DAE.Method/Process]First,this work selects highly-independent journal properties by both correlation coefficient matrix and variance inflation factor;Then,it adopts a progressive strategy to design DAE architectures,analyzes journal distributions generated by four kinds of layers of DAEs and assigns their first latent variables in the latent space as the order metric of journals.Last,our DAE-based ranking is compared with both JCR(Journal Citation Reports)ranking and CI(Academic Journal Clout Index)ranking by constructing a DAE in accordance with balanced partition on journal orders,while compared with CAS(Chinese Academy of Sciences)ranking by constructing another one with imbalanced partition on journal orders respectively.Result/Conclusion]The journals in Library Sciences,Information Science,Law,and Sports are selected as studying objects.The empirical analysis results demonstrate that the proposed approach can not only reveal both global and local relationships between journals in a multiple-level way,but also integrate arbitrary numbers of journal properties in a non-linear way to form one sorting metric for ordering journals.