基于多维特征的引文扩散模式预测研究

Research on Citation Diffusion Pattern Prediction Based on Multidimensional Features

[目的/意义]基于科学论文发表后的早期特征,准确预测论文未来的引文扩散演变模式,对科学产出评估、科学突破早期发现等都具有潜在的价值。[方法/过程]归纳总结9种不同的引文扩散演变模式,并基于论文自发表后的早期时序、结构和文献特征,建模预测未来一定引文窗口内的演变模式。选择美国物理学会的引文数据集进行实证研究,探究不同特征组合下引文扩散演变模式的预测效果。[结果/结论]结果显示,时序特征对预测模型的贡献程度最大,同时结构特征和文献特征也起到重要的作用,当融合3个特征时所有预测模型的准确率均超过了80%,证明了本文所选特征的有效性。

[Purpose/Significance]Based on the early features of scientific papers after publication,accurately predicting the future citation diffusion evolving pattern of papers has potential values for scientific output evaluation and early discovery of scientific breakthroughs,etc.[Method/Process]This study summarized 9 different evolving patterns of citation diffusion.And based on the early chronological,structural and literature features after papers publicated,it modeled and predicted the evolving patterns within a certain citation window in the future.The citation dataset of APS was chosen for empirical research,and predictive effect of the citation diffusion evolution mode under different combinations of features was explored.[Result/Conclusion]The results show that the chronological features contribute the most to the prediction model.At the same time,the structural feature and the literature feature also play an important role.When the three features are combined,the precision of all prediction models exceeds 80%,which proves the effectiveness of the features selected in this article.