On popularity quality: Growth and decay phases of publication popularities

Abstract

Within publication digital collections, citation analysis and publication score assignment are commonly used (i) to evaluate the impact of publications (and scientific collections, e.g., journals and conferences), and (ii) to order digital collection search outputs, e.g., Google Scholar. The popular citation-based web page (and, thus, publication) score measure PageRank is criticized for (a) computing only the current (and, thus, time-independent) publication scores, and (b) not taking into account the fact that citation graphs continuously evolve. Thus, the use of PageRank as is results in penalizing recent publications that have not yet developed enough popularity to receive citations. In order to overcome this inherent bias of PageRank and other citation-based popularity measures, Cho et. al. defined Page Quality for a webpage as its popularity after large numbers of web users become aware of it. Page Quality is based on the assumption that popularity evolves over time. In this paper, we (i) experimentally validate that PageRank scores of publications, as they change over time, follow the logistic growth model that often arises in the context of population growth, (ii) model one aspect of researchers' citation behavior in technology-driven fields (such as computer science) where authors tend not to cite old publications, (iii) argue and empirically verify that publication popularity, unlike web page popularity, has two distinct phases, namely, the popularity growth phase and the popularity decay phase, and (iv) extend the popularity growth model developed by Cho et. al. to capture the popularity decay phase. All of our claims are empirically verified using the ACM SIGMOD Anthology digital collection.