MediaWiki Redirect Tools

Description: Tools to to generate a redirect spells dataset from "raw" MediaWiki XML dumps like those published by the Wikimedia foundation.

If you use this software for research, please cite the following paper:

Hill, Benjamin Mako & Shaw, Aaron. (2014) "Consider the Redirect: A Missing Dimension of Wikipedia Research." In Proceedings of the 10th International Symposium on Open Collaboration (OpenSym 2014). ACM Press. doi: 10.1145/2641580.2641616


To use these tools, you will need need to start with a MediaWiki dump file. For Wikimedia Foundation projects, you can download them all from:

Wikis from and other Wikimedia projects all use the same XML format for their projects.

In the examples in this README, I will use a dump of Simple English Wikipedia that I downloaded with the following command:


Before you start, you may also want to change the default directories for writing intermediate output files. The default directories for writing and reading files are at the top of the file redirect_tools.R and can be changed by editing that file. By default, all files will be written to the subdirectory "./output" in the local directory. If you want to use the default directories, you will still need to create them with a command like this:

mkdir output output/redir output/spells

Step 1: Find Redirects in Revisions




You will run the script to build a dataset of revisions or edits that marks every revisions as either containing a redirect, or not. takes a MediaWiki dump file on STDIN and output a TSV file on STDOUT of the following form: page.title timestamp deleted redirect target
1935456 1935456 17563584 22034930 Mikhail Alekseevich Lavrentiev Mikhail Alekseevich Lavrentiev 1116962833 1125245577 FALSE FALSE FALSE TRUE NA Mikhail Lavrentyev

In this (example) case, the first revision of the article "Mikhail Alekseevich Lavrentiev" was not a redirect but the second is a redirect to "Mikhail Lavrentyev."

If you are using the Simple English dump (which is a single file) you would run the following command to uncompress the dump, parse it using our script, compress the output again, and save the output to the default destination:

7za x -so simplewiki-20140410-pages-meta-history.xml.7z |
python2.7 | bzip2 -c - > output/redir/simple_redirs.tsz.bz2

Because our dumpfile is 7z compressed, I used 7za to uncompress it. If I had used a gzip or bzip compressed file, I would use zcat or bzcat instead. I'm also catting the output to bzip2 -c which will bzip the TSV output to conserve space. The next step assumes a bzip2 compressed file. If you don't want to use bzip2 to compress, you'll need to modify the code.

Step 2: Generate spells




The file redirect_tools.R contains an R function generate.spells() that takes a data frame of edit data as created in step 1 and a list of page titles in order to create a list of redirect spells for those pages. It also contains a function which takes the filename of a bzip compressed file of the form created in step 1 and outputs a full list of redirect spells.

You can run the command with:

R --no-save < 02-generate_spells.R

By default, output will be saved into output/spells. The script will save three versions of the output:

  1. redirect_spells.RData — An RData file suitable for use in GNU R
  2. redirect_spells.tsv — A tab seperated values file suitable for use in a variety of different programs.
  3. redirect_spells.dta — A DTA file suitable for use in Stata (many versions will crop very long artiicle titles due to limitations in the DTA format).

Running Code in Parallel

Because the full history dumps from the WMF foundation are split into many files, it is usually appropriate to parse these dumps in parallel. Although the specific ways you choose to do this will vary by the queuing or scheduling system you use, we've included examples of the scripts we used with Condor on the Harvard/MIT Data Center (HMDC) in the examples/ directory of the source code. They will not work without modification for your computing environment because they have configuration options and paths for our environment hardcoded. That said, they may give you an idea of where you might want to start.

In this parallel code there is a third file 03-assemble_redirect_spells.R that contains R code that will read in all of the separate RData files created in paralell processing, assemble the many smaller dataframes into a single dataframe, and then saves that unified data.frame into a single RData file.