The conventional method of ranking publications is based on the total number of citations received. There are two obvious problems with this method. Firstly, this treats all citations as equal, ignoring differences in importance of citing paper. Secondly, newer publications suffer in relation to older ones because they have simply not been around long enough to achieve a comparable number of citations.
Direct application of Pagerank to citation networks captures the relative importance of citing papers in a manner that is self-consistent. However, because citations cannot be updated after publication, links in citation networks may not represent an up-to-date reflection of relevancy. For an older publication that is not likely to receive any new citations, this is of little consequence. For this reason, the Pagerank of such a publication may be interpreted as its lifetime achievement award. For recent publications, however, the interpretation is not so clear.
Traffic flows along citation links, traveling backwards through time to older papers. Pagerank distributes initial traffic equally across the network. A recent paper therefore has a significantly lower proportion of incoming traffic, due to the smaller number of papers published afterwards.
The assumption, implicit to Pagerank, that traffic distribution is blind to the age of publications is not valid for citation networks. After all, scientific research is time-critical and often builds upon recent progress. Researchers do not usually begin their search by looking at age-old publications, so why model them that way?
The most appropriate ranking method incorporates a more believable traffic distribution for researchers exploring the citation network. A random researcher selects a recent paper with probability that exponentially decreases according to the age of the publication. With a certain probability, the researcher randomly follows one of the citations to the next paper, and so forth.