Abstract
Traditional bibliometric measures often fail to distinguish between the sheer volume of citations and the qualitative influence of those citations. This paper explores the application of weighted PageRank algorithms to author citation networks within the field of Information Retrieval (IR). By modeling the network as a graph of authors and citations, the study distinguishes between "popularity" (raw counts) and "prestige" (citations from established experts). Results indicate that network-aware metrics, particularly those emphasizing prestige, are significantly more effective at identifying award-winning researchers than standard metrics like the h-index.
The Popularity vs. Prestige Divide
In academic evaluation, not all citations carry the same weight. A citation from a world-renowned expert signifies a different level of impact than a citation from a student's first publication. This distinction is characterized as popularity versus prestige.
Standard tools like the h-index or total citation counts treat every citation as a uniform vote, effectively measuring popularity. This paper addresses this gap by applying graph theory—specifically PageRank—to model a scholar's standing based on their position within the professional network.
Methodology: Adapting PageRank
The study adapts the Google PageRank algorithm to a network where authors are nodes and citations are edges. The core contribution is the introduction of Weighted PageRank (WPR), which uses two specific weighting vectors:
- PR_c: Weights authors based on their total citation counts.
- PR_p: Weights authors based on their total number of publications.
The researchers also tested various damping factors ($d$). A lower $d$ (e.g., 0.15) introduces more randomness, while a higher $d$ (e.g., 0.85) emphasizes the actual network topology and the "path" of citations. These results were compared against traditional measures and a specific "Prestige Rank," defined by citations received from other highly-cited papers.
Dataset and Field Growth
The study focused on the field of Information Retrieval (IR), utilizing data from the Web of Science (WOS). The dataset comprised 15,370 papers and 341,871 citations.
To track changes over time, the timeframe (1956–2008) was divided into four chronological phases. As shown in the figure above, the volume of publications in the IR field saw exponential growth in the final decade of the study. To validate the metrics, the researchers used winners of the Gerard Salton Award and the Tony Kent Strix Award as a "ground truth" for prestige.
Key Insights: Network Topology and Influence
The analysis revealed several critical findings regarding how we measure scientific impact:
- Distinct Dimensions: Popularity and prestige are correlated but distinct. Trendy topics may garner high popularity from newcomers, while seminal methodological work earns prestige from domain experts.
- The "Rich Get Richer" Effect: When PageRank is weighted by citation counts ($PR_c$), the results correlate highly with raw popularity. Essentially, weighting by citations reinforces existing citation counts rather than uncovering hidden prestige.
- Scale and Resolution: The difference between popularity and prestige is most pronounced at the top of the field (e.g., the Top 20 authors). At this level, network position becomes a much more sensitive indicator of influence than raw counts.
Results: Validating Prestige
The Prestige Rank (citations from highly-cited papers) emerged as the most effective metric for identifying elite researchers. It captured more Salton and Strix award winners in its top rankings than any other measure, including the h-index and standard PageRank.
Principal Component Analysis (PCA) further confirmed that the 13 indicators tested cluster into four distinct dimensions:
- Popularity: Includes popularity rank, $PR_c$, and Journal Impact Factor.
- Standard PageRank: ($PR$).
- Weighted PageRank: ($PR_p$).
- Prestige: Includes Prestige rank and the h-index.
Notably, while the h-index loaded onto the "Prestige" component, it was still less effective than network-aware prestige rankings.
Limitations and Future Directions
The study acknowledges certain limitations, such as the use of homogeneous networks (author-to-author only) and a first-author bias in the citation records. Future research should explore heterogeneous networks—incorporating journals, institutions, and co-authorship data—and temporal dimensions to create a more robust map of scientific influence.
Summary updated