A Dynamical System for PageRank with Time-Dependent Teleportation

David F. Gleich

Ryan A. Rossi

These codes are research prototypes and may not work for you. No promises. But do email if you run into problems.

Download

{: .nobullets}

Setup

Start matlab in the directory where you unzipped the dynamic_pr.tar.gz file

$ matlab
>> setup_paths
>> load('data/wiki-24hours');

This should work on Mac OSX (Lion tested) and Ubuntu linux (10.10 tested) with Matlab R2011a.

>> v = normcols(v);
>> X = dynamic_pagerank(A,v);

See examples.m for additional examples

Please let us know if you run into any issues.

Overview

The package is organized by directory

/
All of the main matlab codes (dynamic_pagerank.m,...)
ranking
dynamic ranking codes and figures
forecasting
simple models for prediction using Dynamic PageRank
clustering
experimental codes for identifying trends and similar vertices
causality
codes for computing Granger causality between vertices
data
graphs, precomputed data, and script files for extracting and parsing page views
web
this information and all the figures

Figures

|Experiment|Description|Figure| |:------------------|:------------------------------------|:------------------| |fluctuating_interest.m | PageRank dynamical system analytical solution | Fig. 2 | |plot_vertex_yxlims.m | PageRank dynamical system analytical solution | Fig. 3 | |ranking/compute_isim.m | The intersection similarity plot | Fig. 5 | |dpr_timeseries.m | Dynamic PageRank time-series plot | Fig. 6-7 | |forecasting/print_preds_table.m | Performance of Dynamic PageRank for prediction | Tab. 3 | |clustering/dpr_clustering.m | Cluster dynamic score trends, vertices w/ similar behavior | Fig. 8 | |causality/prt_causality.m | Granger causality between vertices | Tab. 4 |