Measuring the impact of a publication in a fair way is a significant challenge in bibliometrics, as it must not introduce biases between fields and should enable comparison of the impact of publications from different years. In this paper, we propose a Bayesian approach to tackle this problem, motivated by empirical data demonstrating heterogeneity in citation distributions. The approach uses the a priori distribution of citations in each field to estimate the expected a posteriori distribution in that field. This distribution is then employed to normalize the citations received by a publication in that field. Our main contribution is the Bayesian Impact Score, a measure of the impact of a publication. This score is increasing and concave with the number of citations received and decreasing and convex with the age of the publication. This means that the marginal score of an additional citation decreases as the cumulative number of citations increases and increases as the time since publication of the document grows. Finally, we present an empirical application of our approach in eight subject categories using the Scopus database and a comparison with the normalized impact indicator Field Citation Ratio from the Dimensions AI database.
翻译:在文献计量学中,公平衡量出版物影响力是一项重大挑战,因为这必须避免不同领域间的偏差,并应能比较不同年份出版物的影响力。本文提出一种贝叶斯方法来解决该问题,其动机源于实证数据所显示的引文分布异质性。该方法利用每个领域的引文先验分布来估计该领域的期望后验分布,随后用此分布对领域内出版物获得的引文进行标准化。我们的主要贡献是贝叶斯影响力得分(Bayesian Impact Score),这是一种衡量出版物影响力的指标。该得分随获得的引文数量增加而呈递增且凹函数特征,随出版物年龄增加而呈递减且凸函数特征。这意味着:随着累计引文数量增加,单次额外引文的边际得分会递减;而当文档发表时间间隔增长时,该边际得分则会递增。最后,我们利用Scopus数据库在八个学科类别中展示了该方法的实证应用,并与Dimensions AI数据库中的标准化影响力指标"领域引文比率"(Field Citation Ratio)进行了比较。