【作者】 赵楠；

【导师】 段宇锋；

【作者基本信息】 华东师范大学 ， 情报学， 2019， 硕士

【摘要】 被引频次是学术论文评价中应用最广泛的指标,与此同时,学者对网络平台论文的下载、点击,等其他效用指标的关注也越来越多,WoS平台发布的的学术用量级指标Usage,在论文评价领域提出了新视角。本研究从论文作者、发表的期刊、所在研究机构等外在因素和论文篇幅、作者数量、参考文献数量等内在因素两方面对比分析高低Usage论文的异同,并尝试通过机器学习的方法实现高Usage论文的预测,以探索高Usage论文的成因。本研究以WoS数据库2013年发表的“COMPUTER SCIENCE ARTIFICIAL INTELLIGENCE”学科下的11008篇论文的题录信息作为原始数据;按照Usage指标降序排序,选取前5%共550篇论文作为高Usage论文集;下载全记录格式的题录信息作为高Usage论文数据集,另外按照发布时间排序取并排除掉高Usage论文的550篇文献,下载全记录格式的题录信息,作为低Usage论文数据集。全部数据下载时间为2018年11月,最终获得1100篇论文的题录信息作为原始数据,然后根据作者机构和期刊的信息,分别在WoS平台的ESI数据库和JCR年度报告中获取论文的机构和期刊的指标数据。研究发现:(1)论文自身特征:两种论文合著率均比较高,高Usage论文的参考文献数量明显大于低Usage论文,而作者数量、论文篇幅,两种论文集不存在显著差异。(2)论文作者:无论是作者的发文量、h指数、还是总被引频次,低Usage论文作者相比于高Usage论文作者,明显处于弱势。(3)论文所在机构:两种论文所在机构在发文量、总被引频次、被引频次/发文量上均不存在显著性差异。这一点与论文作者量化指标不同,作者所在机构的影响力对论文Usage的影响远没有上升到显著水平。(4)高Usage论文所在期刊的总被引频次明显高于低Usage论文所在期刊;在被引半衰期和引用半衰期上高Usage论文也略高一些,但影响因子JIF、发文量、以及文章影响值两种论文不存在显著差异。结果表明,高Usage论文主要集中在,参考文献多,研究基础扎实,作者影响力大,期刊影响力大,引用较新,老化速度较慢的期刊。(5)对比发现CHAID决策树分类模型对Usage指标预测效果最好,模型匹配度较高,在测试集上预测准确率都达到84%以上。更多还原

【Abstract】 Citation frequency is the most widely used index in the evaluation of academic papers.At the same time,scholars pay more and more attention to the download,click and other utility indexes of online papers.The Usage of academic Usage index published by WoS platform puts forward a new perspective in the field of paper evaluation.This study compared and analyzed the external factors such as the author,the published journal,the research institution,and the internal factors such as the length of the paper,the number of authors,and the number of references of the high Usage paper,and tried to predict the high Usage paper through machine learning,so as to explore the causes of high Usage paper.The original data of this study are the quotations of 11008 papers published under the subject of "COMPUTER SCIENCE ARTIFICIAL INTELLIGENCE" in the WoS database in 2013.According to descending order of Usage index,a total of 550 papers in the top 5% were selected as the collection of high Usage papers.The title information in the full record format was downloaded as the data set of high Usage paper.In addition,550 literatures of high Usage paper were taken out in order of release time,and the title information in the full record format was downloaded as the data set of low Usage paper.All data will be downloaded in November 2018,and the bibliographic information of 1100 papers will be obtained as the original data.Then,according to the author’s institution and journal information,the index data of the author’s institution and journal will be obtained in WoS ESI database and JCR annual report respectively.The research found that:(1)The characteristics of the paper itself: the co-authored rate of the two kinds of papers were relatively high,the number of references of the high Usage paper was significantly greater than that of the low Usage paper,while the number of authors,the length of the paper,there was no significant difference between the two kinds of paper collections.(2)The author the paper: no matter the number of articles issued by the author,h index,or the total cited frequency,the author of low Usage paper is obviously in a weak position compared with the author of high Usage paper.(3)The institution of the paper: there is no significant difference in the publication amount,total cited frequency,cited frequency/published amount between the two institutions.What is different from the quantitative index of the author of the paper.The influence of the author’s institution on the value of the paper has not risen to a significant level.(4)The total citation frequency of journals with high Usage was significantly higher than that of journals with low Usage.The paper of high Usage was also slightly higher in cited half-life and cited half-life,but there was no significant difference in the impact factor JIF,the amount of publication,and the impact value of the paper.The results showed that the high Usage paper mainly focused on the journals with many references,solid researc h foundation,great influence of the author,great influence of the journal,relatively new citation and slow aging speed.(5)the comparison shows that the classification model of decision tree has the best prediction effect on Usage index,and the model has a high degree of matching,and the prediction accuracy in the test set is more than 87%.更多还原