影响期刊影响因子的BP神经网络模型构建与实证研究

基于神经网络方法,建立了期刊影响因子与相关影响因素之间非线性对应关系的分析模型,模型以刊期、总被引频次、即年指标、载文量、被引半衰期、引用半衰期、他引比、基金论文比、Web即年下载率等为网络的输入,以影响因子作为网络的输出,利用《中国期刊综合引证报告》提供的数据,进行网络训练和测试,并以有关因素的平均值为基准,采取单因素分析的思想,让一个因素变化其他因素保持平均值,带入训练好的网络进行仿真,观察其输出(影响因子)的变化。分析表明:影响因子受他引比影响最大,受刊期、载文量影响较大,受基金论文比、即年指标影响很小,与Web即年下载率几乎无关。

Analysis Model and Case Study of Relationship between Economic Journals Impact Factor and Related Indexes Based on BP Neural Network

Based on neural network,the paper establishes analysis model about the nonlinear relationship between journal impact factor and related indexes in the process of assessment.The model uses journals publication frequency,the total cited frequency,immediacy index,published paper amount,cited half-life,citing half-life,ratio of other citations,ratio of funded papers and web immediacy download rate as the inputs,and takes impact factor as the output.Using the datum of Chinese Academic Journals Comprehensive Citation Report,network training and testing are applied.At the same time,it takes the averages of the related factors as the benchmark and utilizes the idea of single-factor analysis to change one of the factors' value while others remain their averages.The changed datum are used into the trained network to simulate the outputs(impact factor),in order to make further studies.The results show that impact factor is mostly influenced by ratio of other citations,less by journals publication frequency and published paper amount,and least by ratio of funded papers and immediacy index,and nearly irrespectively by web immediacy download rate.