Ranking the Future Influence of Scientific Literatures

The future influence of scientific literatures plays an important role under many decision-making circumstances like choosing which recently published papers to follow or identifying rising stars from different scientific research domains. But the large number of scientific literatures published every year disables researchers to manually evaluate the quality and potential of all the literatures. Therefore, the problem of automatically ranking the future influence of scientific literatures has drawn much interest. Previous works have typically addressed this problem by incorporating heuristically designed time-aware functions into traditional graph-based ranking methods. These functions were developed by manually analyzing the target datasets and aimed to model the underlying dynamic nature. In contrast, there is no generic method which can adaptively capture the dynamic nature of different datasets without manual intervention. In this paper, we focus on addressing the issue of ranking the future influence of scientific literatures. Our main contribution is a generic and effective method which adaptively learns the underlying dynamic nature of different scientific literature datasets and applies the learned knowledge to ranking. This method creatively transforms the raw problem into a learning to rank problem with the help of HSHMRR, a new mutual reinforcement ranking framework used to precisely measure the importance of different types of scientific entities (papers, researchers, venues and institutions) in different time periods. Compared with previous works, our proposed method can be directly applied on different datasets with different target time periods and different definition of the future influence. Experiment results on three datasets extracted from Microsoft Academic Graph confirm the effectiveness of our proposed method, which outperforms the state-of-the-art methods by at most 29% in terms of Spearman's rank correlation coefficient.