1 MediaWiki Redirect Tools
2 =======================================================================
4 | **Author:** `Benjamin Mako Hill`__ <mako@atdot.cc> and `Aaron Shaw`__ <aaron.d.shaw@gmail.com>
5 | **Homepage:** http://networkcollectiv.es/wiki-redirects/
6 | **Code:** http://projects.mako.cc/source/?p=redirect-tools
7 | **License:** `GNU GPLv3 or any later version`__
8 | **Description:** Tools to to generate a redirect spells dataset from "raw" MediaWiki XML dumps like those published by the Wikimedia foundation.
11 __ http://aaronshaw.org/
12 __ http://www.gnu.org/copyleft/gpl.html
14 If you use this software for research, please cite the following
17 *Hill, Benjamin Mako and Aaron Shaw. "Consider the Redirect: A
18 Missing Dimension of Wikipedia Research." In Proceedings of the 10th
19 International Symposium on Open Collaboration (OpenSym 2014). ACM
24 To use these tools, you will need need to start with a MediaWiki dump
25 file. For Wikimedia Foundation projects, you can download them all from:
26 http://dumps.wikimedia.org/
28 Wikis from Wikia.com and other Wikimedia projects all use the same XML format
31 In the examples in this README, I will use a dump of `Simple English
32 Wikipedia`__ that I downloaded with the following command::
34 wget http://dumps.wikimedia.org/simplewiki/20140410/simplewiki-20140410-pages-meta-history.xml.7z
36 __ https://simple.wikipedia.org/
38 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::
40 mkdir output output/redir output/spells
42 Step 1: Find Redirects in Revisions
43 -----------------------------------------
48 - Wikimedia Utilities (https://bitbucket.org/halfak/wikimedia-utilities)
52 - Wikimedia XML Dump files (compressed in some form)
56 - bzip2 compressed TSV files (one line per revision)
58 You will run the `01-extract_redirects.py` script to build a dataset of revisions or edits that marks every revisions as either containing a redirect, or not. `01-extract_redirects.py` takes a MediaWiki dump file on STDIN and output a TSV file on STDOUT of the following form:
60 +---------+-------------+--------------------------------+------------+---------+----------+--------------------+
61 | page.id | revision.id | page.title | timestamp | deleted | redirect | target |
62 +=========+=============+================================+============+=========+==========+====================+
63 | 1935456 | 17563584 | Mikhail Alekseevich Lavrentiev | 1116962833 | FALSE | FALSE | NA |
64 | 1935456 | 22034930 | Mikhail Alekseevich Lavrentiev | 1125245577 | FALSE | TRUE | Mikhail Lavrentyev |
65 +---------+-------------+--------------------------------+------------+---------+----------+--------------------+
68 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."
70 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::
72 7za x -so simplewiki-20140410-pages-meta-history.xml.7z |
73 python2.7 01-extract_redirects.py | bzip2 -c - > output/redir/simple_redirs.tsz.bz2
75 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.
78 Step 2: Generate spells
79 -----------------------------------------
84 - data.table (http://cran.r-project.org/web/packages/data.table/)
85 - foreign (http://cran.r-project.org/web/packages/foreign/)
89 - bzip compressed TSV files
93 - RData files containing a data.frame of redirect spells named `redirect.spell`
94 (one file per input file)
95 - Stata DTA file (same data)
96 - TSV file (same data)
98 The file `redirect_tools.R` contains an R function `generate.spells()` that
99 takes a data frame of edit data as created in step 1 and a list of page titles
100 in order to create a list of redirect spells for those pages. It also
101 contains a function `filename.to.spells()` which takes the filename of a bzip
102 compressed file of the form created in step 1 and outputs a full list of
105 You can run the command with::
107 R --no-save < 02-generate_spells.R
109 By default, output will be saved into `output/spells`. The script will
110 save three versions of the output:
112 1. `redirect_spells.RData` — An RData file suitable for use in GNU R
113 2. `redirect_spells.tsv` — A tab seperated values file suitable for use in a variety of different programs.
114 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).
117 Running Code in Parallel
118 -----------------------------------------
120 Because the full history dumps from the WMF foundation are split into
121 many files, it is usually appropriate to parse these dumps in
122 parallel. Although the specific ways you choose to do this will vary
123 by the queuing or scheduling system you use, we've included examples
124 of the scripts we used with Condor on the Harvard/MIT Data Center
125 (HMDC) in the `examples/` directory of the source code. They will not
126 work without modification for your computing environment because they
127 have configuration options and paths for our environment
128 hardcoded. That said, they may give you an idea of where you might
131 In this parallel code there is a third file
132 `03-assemble_redirect_spells.R` that contains R code that will read in
133 all of the separate RData files created in paralell processing,
134 assemble the many smaller dataframes into a single dataframe, and then
135 saves that unified data.frame into a single RData file.