Pruning dangling nodes before computing Pagerank when your document
corpus is very large does not change the individual Pagerank of the
remaining nodes. However, dangling nodes may have very high Pagerank
due to a large number of in-links. After computing an approximate
Pagerank for the corpus, you can return the nodes to the graph, and
give them a portion of the Pagerank of their parent nodes. The
differences are easily noticed with small graphs, but when millions of
nodes are considered, the approximation is closer to standard Pagerank
Kalin Jonas
