Realistic Citation Count Prediction Task for Newly Published Papers

Findings of the Association for Computational Linguistics: EACL 2023, pages 1131–1141 May 2-6, 2023 ©2023 Association for Computational Linguistics Realistic Citation Count Prediction Task for Newly Published Papers Jun Hirako Ryohei Sasano Koichi Takeda Graduate School of Informatics, Nagoya University hirako.jun.e5@s.mail.nagoya-u.ac.jp {sasano,takedasu}@i.nagoya-u.ac.jp Abstract Citation count prediction is the task of predict- ing the future citation counts of academic pa- pers, which is particularly useful for estimat- ing the future impacts of an ever-growing num- ber of academic papers. Although there have been many studies on citation count prediction, they are not applicable to predicting the cita- tion counts of newly published papers, because they assume the availability of future citation counts for papers that have not had enough time pass since publication. In this paper, we first identify problems in the settings of existing studies and introduce a realistic citation count prediction task that strictly uses information available at the time of a target paper’s publica- tion. For realistic citation count prediction, we then propose two methods to leverage the cita- tion counts of papers shortly after publication. Through experiments using papers collected from arXiv and bioRxiv, we demonstrate that our methods considerably improve the perfor- mance of citation count prediction for newly published papers in a realistic setting. 1 Introduction In recent years, the number of academic papers in various fields has increased drastically. Accord- ingly, the demand for techniques for predicting papers that will become influential in the future is growing to help readers identify those papers and support efficient knowledge acquisition. In this study, we adopt the citation count as a measure of future impact, following several previous studies (e.g., Chubin and Garfield, 1979; Aksnes, 2006), and we address the citation count prediction task, which entails predicting how many times a target paper will be cited in the future. There have been many studies on citation count prediction (e.g., Fu and Aliferis, 2008; van Dongen et al., 2020). However, none of those settings is strictly applicable to predicting the citation count of newly published papers, because they assume Citation count one year after publication Papers published more than 12 months ago Papers published 6 months ago Papers published 3 months ago Target paper Figure 1: Comparison of a realistic citation count pre- diction setting with existing research settings. Each bar (■+□) represents the citation count one year after pub- lication, which existing studies assume to be available, while the gray part (■) represents the citation count that is actually available at the time of a target paper’s publication. the availability of future citation counts for papers shortly after publication. For example, consider the case of predicting the citation counts one year after publication. For training and testing, the cor- rect citation count of the target paper one year after publication must be known; hence, only papers published more than one year ago are used in the experiments. Consequently, even for papers pub- lished less than one year before the target paper, the number of citations one year …

Content truncated to 500 words