基于知识关联度的科学论文扩散效果预测研究——早期施引文献的作用  

Predicting Scientific Paper Diffusion Effect Based on Knowledge Association——The Role of Early Citing Publications

作　　者：李悦[1,2] 马亚雪 张宇 孙建军[1,2] Li Yue;Ma Yaxue;Zhang Yu;Sun Jianjun(Laboratory of Data Intelligence and Interdisciplinary Innovation,Nanjing University,Nanjing 210023,China;School of Information Management,Nanjing University,Nanjing 210023,China)

机构地区：[1]南京大学数据智能与交叉创新实验室,江苏南京210023 [2]南京大学信息管理学院,江苏南京210023

出　　处：《现代情报》2023年第11期73-84,共12页Journal of Modern Information

基　　金：国家自然科学基金青年项目“基于特征挖掘的科学问题域创新状态建模与突破机理研究”(项目编号:72204109);中国博士后科学基金项目“基于状态特征挖掘的科学突破影响机制与激发模型研究”(项目编号:2022M711568)。

摘　　要：[目的/意义]基于早期施引文献与科学论文的知识关联对科学论文扩散效果进行预测,有助于从价值反馈角度前瞻性识别高影响力学术论文,为科研人员建立科学研究成果早期学术影响力评估体系提供参考。[方法/过程]测度早期施引文献与目标科学论文在主题、期刊和作者3个层面的关联程度,采用线性回归与负二项回归模型,挖掘3种类型的知识关联度与目标科学论文扩散效果(即扩散速度、广度和强度)的内在关联机制;在此基础上引入机器学习算法对科学论文的扩散效果进行预测,剖析3类知识关联特征在预测任务中的重要性排序。[结果/结论]神经科学领域的实证分析显示,主题关联与目标科学论文的扩散速度呈正相关关系,与扩散广度和扩散强度呈倒U型关系;期刊关联会抑制目标科学论文的扩散速度,但能够正向影响其扩散强度与扩散广度;作者关联仅对扩散强度有稳定的正向影响;基于主题关联与期刊关联可以实现对科学论文扩散速度的有效预测,但难以预测扩散广度和扩散强度。随机森林模型在扩散速度预测中性能最佳,主题关联特征的重要性高于期刊关联。[Purpose/Significance]This study predicts the diffusion effect of literature based on the knowledge association between target scientific papers and their early citing publications.It helps in the prospective identification of high-impact academic papers from the perspective of value feedback and provides a reference for researchers to establish an early evaluation system for scientific performance.[Method/Process]This study measured the degree of association between target scientific papers and their early citing publications from three perspectives,i.e.,topic,journal,and author,and adopted the linear regression and negative binomial regression models to dissect the key factors affecting the diffusion effect(i.e.,diffusion speed,breadth,and intensity).Based on the regression result,the study incorporated machine learning algorithms to predict diffusion effect of scientific papers and analyzed the importance ranking of the three types of knowledge association features in the prediction task.[Result/Conclusion]Subject association positively promotes the diffusion speed of scientific literature but shows an inverted U-shaped relationship with the diffusion breadth and intensity.Journal association inhibits the diffusion speed but can positively affect the diffusion intensity and breadth.Author association only has a consistent positive impact on the diffusion intensity.Predicting diffusion speed can be effectively achieved based on topic and journal associations,although accurately predicting diffusion breadth and intensity through knowledge associations proves to be challenging.The random forest model performs best in predicting diffusion speed,with topic association being of higher importance compared to journal association.

关 键 词：扩散效果预测 引文扩散 知识关联度 早期施引文献 

分 类 号：G250.252[文化科学—图书馆学]