# R igraph manual pages

Use this if you are using igraph from R

## The Page Rank algorithm

### Description

Calculates the Google PageRank for the specified vertices.

### Usage

page_rank(graph, algo = c("prpack", "arpack", "power"),
vids = V(graph), directed = TRUE, damping = 0.85,
personalized = NULL, weights = NULL, options = NULL)

page_rank_old(graph, vids = V(graph), directed = TRUE, niter = 1000,
eps = 0.001, damping = 0.85, old = FALSE)


### Arguments

 graph The graph object. algo Character scalar, which implementation to use to carry out the calculation. The default is "prpack", which uses the PRPACK library (https://github.com/dgleich/prpack). This is a new implementation in igraph version 0.7, and the suggested one, as it is the most stable and the fastest for all but small graphs. "arpack" uses the ARPACK library, the default implementation from igraph version 0.5 until version 0.7. power uses a simple implementation of the power method, this was the default in igraph before version 0.5 and is the same as calling page_rank_old. vids The vertices of interest. directed Logical, if true directed paths will be considered for directed graphs. It is ignored for undirected graphs. damping The damping factor (‘d’ in the original paper). personalized Optional vector giving a probability distribution to calculate personalized PageRank. For personalized PageRank, the probability of jumping to a node when abandoning the random walk is not uniform, but it is given by this vector. The vector should contains an entry for each vertex and it will be rescaled to sum up to one. weights A numerical vector or NULL. This argument can be used to give edge weights for calculating the weighted PageRank of vertices. If this is NULL and the graph has a weight edge attribute then that is used. If weights is a numerical vector then it used, even if the graph has a weights edge attribute. If this is NA, then no edge weights are used (even if the graph has a weight edge attribute. This function interprets edge weights as connection strengths. In the random surfer model, an edge with a larger weight is more likely to be selected by the surfer. options Either a named list, to override some ARPACK options. See arpack for details; or a named list to override the default options for the power method (if algo="power"). The default options for the power method are niter=1000 and eps=0.001. This argument is ignored if the PRPACK implementation is used. niter The maximum number of iterations to perform. eps The algorithm will consider the calculation as complete if the difference of PageRank values between iterations change less than this value for every node. old A logical scalar, whether the old style (pre igraph 0.5) normalization to use. See details below.

### Details

For the explanation of the PageRank algorithm, see the following webpage: http://infolab.stanford.edu/~backrub/google.html, or the following reference:

Sergey Brin and Larry Page: The Anatomy of a Large-Scale Hypertextual Web Search Engine. Proceedings of the 7th World-Wide Web Conference, Brisbane, Australia, April 1998.

igraph 0.5 (and later) contains two PageRank calculation implementations. The page_rank function uses ARPACK to perform the calculation, see also arpack.

The page_rank_old function performs a simple power method, this is the implementation that was available under the name page_rank in pre 0.5 igraph versions. Note that page_rank_old has an argument called old. If this argument is FALSE (the default), then the proper PageRank algorithm is used, i.e. (1-d)/n is added to the weighted PageRank of vertices to calculate the next iteration. If this argument is TRUE then (1-d) is added, just like in the PageRank paper; d is the damping factor, and n is the total number of vertices. A further difference is that the old implementation does not renormalize the page rank vector after each iteration. Note that the old=FALSE method is not stable, is does not necessarily converge to a fixed point. It should be avoided for new code, it is only included for compatibility with old igraph versions.

Please note that the PageRank of a given vertex depends on the PageRank of all other vertices, so even if you want to calculate the PageRank for only some of the vertices, all of them must be calculated. Requesting the PageRank for only some of the vertices does not result in any performance increase at all.

Since the calculation is an iterative process, the algorithm is stopped after a given count of iterations or if the PageRank value differences between iterations are less than a predefined value.

### Value

For page_rank a named list with entries:

 vector A numeric vector with the PageRank scores. value The eigenvalue corresponding to the eigenvector with the page rank scores. It should be always exactly one. options Some information about the underlying ARPACK calculation. See arpack for details. This entry is NULL if not the ARPACK implementation was used.

For page_rank_old a numeric vector of Page Rank scores.

### Author(s)

Tamas Nepusz ntamas@gmail.com and Gabor Csardi csardi.gabor@gmail.com

### References

Sergey Brin and Larry Page: The Anatomy of a Large-Scale Hypertextual Web Search Engine. Proceedings of the 7th World-Wide Web Conference, Brisbane, Australia, April 1998.

Other centrality scores: closeness, betweenness, degree

### Examples


g <- sample_gnp(20, 5/20, directed=TRUE)
page_rank(g)$vector g2 <- make_star(10) page_rank(g2)$vector

# Personalized PageRank
g3 <- make_ring(10)
page_rank(g3)$vector reset <- seq(vcount(g3)) page_rank(g3, personalized=reset)$vector


[Package igraph version 1.2.4 Index]