Fusing structural and temporal information in citation networks for identifying milestone works

The rapid proliferation of scientific and technological works has highlighted the necessity to effectively identify the significant achievements in more and more complex citation networks. Mainstream algorithms fall into two categories: structural information based algorithms with Citation and PageRank as the core; Temporal information based algorithms represented by the citation dynamic model and Relevance. This article conducts a detailed study of the relationship between these two categories to fill the gap in this area. We use the American Physical Society (APS) dataset, which includes 469,452 papers and 5,016,382 citations from 1893 to 2010. Our findings indicate that PageRank and Citation are statistically similar, both favoring older articles. However, Relevance excels in early forecasting, hence showing a weaker correlation with PageRank. Inspired by this, we introduce a new method called Structural-Temporal Rank (STRank). Validation experiments demonstrate that STRank excels in identifying milestone letters and predicting future impact, outperforming other methods in these tasks. This study introduces the idea of fusing structural and temporal information in designing ranking methods that could guide the future development of more efficient node identification algorithms in networks.