Promise and Pitfalls of Extending Google's PageRank Algorithm to Citation Networks

TL;DR

Citation counts are widely used to judge scientific importance, but they ignore the influence of citing papers and can understate older work. The paper applies PageRank to APS citations and extends it with CiteRank; PageRank surfaces modestly cited papers, while CiteRank targets relevance to currently popular research directions.

• Problem

  Citation counts are widely used to measure publication importance, but they give equal weight to every citation and can underestimate influential older work.

  Source: p. 1

• Method

  The authors apply PageRank to the APS citation network and modify it as CiteRank by exponentially favoring more recent publications at the start of searches.

  Source: pp. 1–2

• Results

  PageRank outliers include modestly cited, old papers; CiteRank with d = 0.5 and τ = 2.6 years quantifies relevance to currently popular research directions.

  Source: pp. 1, 3

• Takeaways & Limitations

  PageRank extends citation counts by weighting citations from important publications, including those with short reference lists.

  Source: pp. 1, 3

  For citation networks, the authors argue for d = 1/2, corresponding to a citation chain of two links.

  Source: p. 1