WP2TXT

Wikipedia dump file to text converter

IMPORTANT: This is a project still work in progress and it could be slow, unstable, and even destructive! It should be used with caution.

About

WP2TXT extracts plain text data from Wikipedia dump file (encoded in XML/compressed with Bzip2) stripping all the MediaWiki markups and other metadata. It is originally intended to be useful for researchers who look for an easy way to obtain open-source multi-lingual corpora, but may be handy for other purposes.

UPDATE: Version 0.9.1 has added a new option num-threads, which improves the performance significantly . Note also that --category option is enabled by default, resulting with output format somewhat different from previous versions. Check out the new format using test data in data/output_samples folder before going on to convert a huge wikipedia dump.

Features

Convert dump files of Wikipedia of various languages (I hope).
Create output files of specified size.
Allow users to specify text elements to be extracted/converted (page titles, section titles, lists, and tables).

Installation

$ gem install wp2txt

Usage

Obtain a Wikipedia dump file (from here) with a file name such as:

xxwiki-yyyymmdd-pages-articles.xml.bz2

where xx is language code such as "en (English)" or "ja (Japanese)", and yyyymmdd is the date of creation (e.g. 20120601).

Command line options are as follows:

Usage: wp2txt [options]
where [options] are:
           --input-file, -i:   Wikipedia dump file with .bz2 (compressed) or
                               .txt (uncompressed) format
       --output-dir, -o <s>:   Output directory (default: current directory)
--convert, --no-convert, -c:   Output in plain text (converting from XML)
                               (default: true)
      --list, --no-list, -l:   Show list items in output (default: true)
--heading, --no-heading, -d:   Show section titles in output (default: true)
    --title, --no-title, -t:   Show page titles in output (default: true)
                --table, -a:   Show table source code in output
               --inline, -n:   leave inline template notations unmodified
            --multiline, -m:   leave multiline template notations unmodified
                  --ref, -r:   leave reference notations in the format
                               [ref]...[/ref]
             --redirect, -e:   Show redirect destination
  --marker, --no-marker, -k:   Show symbols prefixed to list items,
                               definitions, etc. (Default: true)
             --category, -g:   Show article category information
        --file-size, -f <i>:   Approximate size (in MB) of each output file
                               (default: 10)
      -u, --num-threads=<i>:   Number of threads to be spawned (capped to the number of CPU cores; 
                               set 99 to spawn max num of threads) (default: 4)
              --version, -v:   Print version and exit
                 --help, -h:   Show this message

Caveats

Certain types of data such as mathematical equations and computer source code are not be properly converted. Please remember this software is originally intended for correcting “sentences” for linguistic studies.
Extraction of normal text data could sometimes fail for various reasons (e.g. illegal matching of begin/end tags, language-specific conventions of formatting, etc).
Conversion process can take far more than you would expect. It could take several hours or more when dealing with a huge data set such as the English Wikipedia on a low-spec environments.
Because of nature of the task, WP2TXT needs much machine power and consumes a lot of memory/storage resources. The process thus could halt unexpectedly. It may even get stuck, in the worst case, without getting gracefully terminated. Please understand this and use the software at your own risk.

Useful Link

Wikipedia Database backup dumps

Author

Yoichiro Hasebe (yohasebe@gmail.com)

References

The author will appreciate your mentioning one of these in your research.

Yoichiro HASEBE. 2006. Method for using Wikipedia as Japanese corpus. Doshisha Studies in Language and Culture 9(2), 373-403.
長谷部陽一郎. 2006. Wikipedia日本語版をコーパスとして用いた言語研究の手法. 『言語文化』9(2), 373-403.

License

This software is distributed under the MIT License. Please see the LICENSE file.

yukishimazu/wp2txt