基于神经网络的国内呼吸病学、结核病学核心期刊论文热度预测初探  

Preliminary study on the predicting of the popularity of papers in the core journals of respiratory disease and tuberculosis in China based on neural network

作　　者：郭萌 朱玉辉 范永德 李敬文 Guo Meng;Zhu Yuhui;Fan Yongde;Li Jingwen(Chinese Journal of Antituberculosis Publishing House,Beijing 100035,China;School of Statistics,Renmin University of China,Beijing 100872,China)

机构地区：[1]《中国防痨杂志》期刊社,北京100035 [2]中国人民大学统计学院,北京100872

出　　处：《结核与肺部疾病杂志》2025年第1期79-86,共8页Journal of Tuberculosis and Lung Disease

摘　　要：目的:在神经网络这一基于统计学的模型广泛应用于微博、微信公众号热度预测的背景下,探索神经网络应用于学术论文领域热度预测的价值,为评价学术论文水平提供一种新的辅助检测手段。方法:以2019-2021年国内“呼吸病学、结核病学”核心期刊(《中国防痨杂志》《中华结核和呼吸杂志》《临床肺科杂志》《中国呼吸与危重监护杂志》《国际呼吸杂志》《中华肺部疾病杂志(电子版)》)已发表的论文作为样本,采取蕴含文章重要信息的标题、摘要、关键词及文章发表天数作为神经网络的输入层,并通过分词工具对输入层进行分词获取特征,进而预测文章的被引用量。结果:2019-2021年,“呼吸病学、结核病学”核心期刊共计发表论文4729篇,2019-2021年各年分别发表论文1690、1534、1505篇。根据被引频次的分类,“呼吸病学、结核病学”核心期刊2019-2021年高被引论文(被引频次30~250次)、中被引论文(被引频次4~29次)、低被引论文(被引频次1~3次)、零被引论文分别为46、1362、1872、1449篇,分别占0.97%、28.80%、39.59%、30.64%。通过神经网络分析,对文章被引用量预测的准确率、精确率和召回率分别达到99.68%、99.63%和99.65%。结论:作为人工智能技术的一种方法,神经网络可逐步引入到学术领域,为客观公正地评判稿件水平提供一种辅助的检测手段,以弥补编辑初审和外审存在的局限性。s,keywords and publication days of papers from the domestic core journals of“Respiratory disease and tuberculosis”(Chinese Journal of Antituberculosis,Chinese Journal of Tuberculosis and Respiratory Diseases,Journal of Clinical Pulmonary Medicine,Chinese Journal of Respiratory and Critical Care Medicine,International Journal of Respiration,and Chinese Journal of Lung Diseases(Electronic Edition))were collected as the input layer of the neural network from 2019 to 2021,and the features was obtained using segmentation tools to segment the input layer,thereby the citation count of the papers was predicted.Results:From 2019 to 2021,a total of 4729 papers were published in the core journals of“Respiratory disease and tuberculosis”,and 1690,1534 and 1505 papers were published in each year from 2019 to 2021.According to the classification of citation frequency,from 2019 to 2021,the number of highly cited papers(30 to 250 times),moderately cited papers(4 to 29 times),low cited papers(1 to 3 times),and zero cited papers in the core journals of“Respiratory disease and tuberculosis”were 46,1362,1872,and 1449,respectively,accounting for 0.97%,28.80%,39.59%,and 30.64%.Using neural networks,the accuracy,precision,and recall of predicting article citation rates had reached 99.68%,99.63%,and 99.65%,respectively.Conclusion:As a method of artificial intelligence technology,neural networks could be gradually introduced into the academic field to provide an auxiliary detection method for objectively and fairly evaluating the level of manuscripts,in order to compensate for the limitations of initial and external review by editors.

关 键 词：神经网络(计算机) 期刊论文 预测 

分 类 号：G255.2[文化科学—图书馆学] TP183[自动化与计算机技术—控制理论与控制工程]