Source-linked AI summary
Applying weighted PageRank to author citation networks
Ying Ding
TL;DR
Prior citation-count measures do not distinguish citations by citing author or account for network features of citing behavior, so the paper tests weighted PageRank on author citation networks. It reports significant correlation between popularity and prestige ranks across four phases and finds prestige rank outperformed other indicators in identifying prize winners.
Problem
Earlier citation-count measures do not distinguish citations by citing author or account for network features of citing behavior.
Source: Related Work · p. 5Method
The paper applies weighted PageRank to author citation networks using citation and publication as weighted vectors.
Source: Data Collection · p. 8Results
Popularity and prestige ranks were significantly correlated across all four phases, and prestige rank outperformed other indicators across top-rank categories in identifying IR prize winners.
Source: Evaluation · p. 15, Conclusion · p. 16Takeaways & Limitations
For identifying prize winners in IR, the paper concludes that prestige rank is preferable to other measures.
Source: Evaluation · p. 15
Abstract
from arXiv · showhide
This paper aims to identify whether different weighted PageRank algorithms can be applied to author citation networks to measure the popularity and prestige of a scholar from a citation perspective. Information Retrieval (IR) was selected as a test field and data from 1956-2008 were collected from Web of Science (WOS). Weighted PageRank with citation and publication as weighted vectors were calculated on author citation networks. The results indicate that both popularity rank and prestige rank were highly correlated with the weighted PageRank. Principal Component Analysis (PCA) was conducted to detect relationships among these different measures. For capturing prize winners within the IR field, prestige rank outperformed all the other measures.
Introduction
This section examines whether author-level PageRank variants, including weighted PageRank, can represent popularity and prestige in citation networks, using Information Retrieval author-citation data. It describes the data and ranking measures, their relationships to popularity and prestige, and their performance in capturing IR prize winners.
Introduction
- Popularity reflects total citations, whereas prestige reflects citations from highly cited publications; researchers can score highly on one without the other.The introduction illustrates the distinction between broad citation and recognition by domain experts.Source: Introduction · p. 2
- The paper tests whether PageRank and weighted PageRank on author citation networks can represent scholars’ popularity and prestige.Information retrieval is the test field, and the study compares the measures with citation counts, the h-index, and Ding and Cronin’s popularity and prestige ranks.Source: Introduction · p. 4
- Using 15,370 information-retrieval papers and 341,871 citations from 1956–2008, the study compared measures, including weighted PageRank, by award-winner counts in top 5 and top 10 lists.Source: Introduction · p. 4
Related Work
- Prior citation-count measures do not distinguish citing authors or account for citation-network features, while weighted PageRank has been applied to journal and author networks.Examples include author citation-network approaches by Radicchi, Fortunato, Makines and Vespignani (2009).Source: Related Work · pp. 5–6
- Prior work defined AuthorRank as a PageRank modification that considers link weights among co-authorships.Source: Related Work · p. 6
Data Collection
- The study used 15,370 IR papers and 341,871 citations from Web of Science (1956–2008) to build author citation networks and calculate PR, PR_c, and PR_p.The networks represented authors as nodes and citing relationships as weighted directed edges; citation counts or publication counts served as weighted vectors.Source: Data Collection · pp. 6–8
Top 20 Ranks
- Across four phases, prestige rank differed dramatically from popularity rank and PageRank for the top 20, while popularity and prestige ranks were significantly correlated for top 100 highly cited authors.The authors conclude that ranking scale matters.Source: Top 20 Ranks · p. 9, Top 100 Ranks · p. 12
Top 100 Ranks
- Popularity rank correlated most with PR_c(.15) (mean r=0.931, p<0.01); prestige rank correlated more with PR_c than with PR or PR_p.Popularity correlations were PR_c (mean r=0.925, p<0.01), PR (mean r=0.702, p<0.01), and PR_p (mean r=0.485, p<0.01); prestige correlations were PR_c (mean r=0.616, p<0.01), PR (mean r=0.379, p<0.01), and PR_p (mean r=0.588, p<0.01), with the highest correlation specifically for PR_c(.85) in Phase 2 and Phase 3.Source: Top 100 Ranks · pp. 10–11, Table 2 · p. 11
Correlation of various indicators
- In 2001–2008, prestige rank was highly correlated with PR_p(.85), while impact-factor rank correlated more with popularity than prestige rank.The h-index rank was significantly correlated with popularity rank, prestige rank, PR_c, PR_p(.85), or PR_p(.5).Source: Table 3 · p. 13, Correlation of various indicators · p. 13
- The PCA extracted four components explaining 87.84% of total variance using Varimax with Kaiser Normalization.Source: Correlation of various indicators · p. 14, Table 4 · p. 14
- PCA distinguished popularity (popularity rank, PR_c, IF rank), PageRank (PR, PR_c), weighted PageRank (PR_p), and prestige (prestige rank, h-index rank, IF rank) dimensions.Popularity rank and prestige rank belonged to different components.Source: Correlation of various indicators · p. 14, Table 4 · p. 14
Evaluation
- Prestige rank outperformed all other indicators for capturing IR prize winners across the top 5, 10, 20, and 50 ranks in the 2001–2008 dataset.The prize set included the Gerard Salton Award and Tony Kent Strix Award.Source: Evaluation · p. 15, Table 5 · p. 15
Conclusion · Footnotes
The study applies weighted PageRank to author citation networks and finds that prestige rank best captures prize winners, while popularity and prestige rankings show distinct relationships across analyses. It also points toward multidimensional impact indicators and weighted citations as future methodological directions.
Conclusion
- Using WOS IR data from 1956–2008, the study divided the period into four phases and tested weighted PageRank with citation- or publication-count vectors and damping factors d=0.15 and d=0.85.Source: Conclusion · p. 15
- Across four phases, popularity and prestige ranks were significantly correlated among the top 100, while the top 20 showed discrepancies across ranking measures.The compared measures included prestige rank, popularity rank, PageRank, and weighted PageRank.Source: Conclusion · pp. 15–16
- In the 2001–2008 PCA, popularity rank and prestige rank loaded on separate components, indicating a difference in what they measure.The four extracted components represented popularity, original PageRank, weighted PageRank, and prestige.Source: Conclusion · p. 16
- In 2001–2008, prestige rank outperformed all other indicators in the tested top 5, 10, 20, and 50 ranks for identifying prize winners.The comparison tested 13 measures.Source: Conclusion · p. 16
- The paper frames scholarly-impact evaluation around indicators, counting methods, and datasets, and argues that linking different data types can support new evaluation measures.Source: Conclusion · p. 16
- The conclusion identifies multidimensional impact indicators and weighted citations as future directions, including measures combining publications, citations, and co-authorship.It notes the need to distinguish citations by the quality of the citing paper.Source: Conclusion · p. 16