A temporal evolution and fine-grained information aggregation model for citation count prediction

Scientific papers serve as the primary medium for disseminating scientific knowledge, containing information that advances research progress. The development of automated prediction techniques for citation counts of scientific papers could expedite the identification of valuable contributions within an extensive corpus of literature. Nevertheless, most existing methods neglect the importance of capturing temporal evolution information and fine-grained information aggregation during the acquisition of citation networks. To tackle the aforementioned issues, we propose the Temporal Evolution and Fine-grained Information Aggregation model (TEFIA) for predicting citation counts. The TEFIA model effectively utilizes temporal evolution information from citation networks and seamlessly integrates fine-grained aggregation of information from diverse paper attributes, thereby significantly improving the accuracy of citation count prediction. Specifically, we conceptualize scientific papers as citation networks and introduce a network representation module aimed at acquiring feature representations of nodes. The module for temporal evolution representation captures the temporal evolution features inherent in citation networks. Furthermore, the module for fine-grained information aggregation integrates information from diverse types of attribute nodes associated with scientific papers. Lastly, the citation prediction module forecasts the future citation counts of scientific papers. The TEFIA model is applied to comprehensive experiments on real-world datasets. The proposed model achieves reductions of 9.40 and 6.49% in MALE prediction errors compared to state-of-the-art methods on the APS and AMiner datasets, respectively. This study demonstrates the utilization of temporal evolution representations and fine-grained information aggregation to improve the performance of citation count prediction models.