panx27 / wikiann

Wikimedia dumps processor

Wikipedia

Quickstart

 cd wikiann/wikipedia/dump

Step 1: download wiki dumps

 USAGE: download.py <wikisite (e.g., enwiki)> <label (e.g., 20180401)> <output dir>

 e.g., python download.py yowiki latest ~/wikiann

Step 2: parse wiki dump

 usage: parse_pages_articles_multistream.py [-h] [--nworker NWORKER] [--verbose]
                                            [--index_range INDEX_RANGE]
                                            p_xml p_index outdir

 positional arguments:
   p_xml                 path to pages-articles-multistream.xml.bz
   p_index               path to pages-articles-multistream-index.txt.bz2
   outdir                output directory

 optional arguments:
   -h, --help            show this help message and exit
   --nworker NWORKER, -n NWORKER
                         number of processors to use (default=1)
   --verbose, -v         verbose logging
   --index_range INDEX_RANGE, -i INDEX_RANGE
                         Index range for debug

 e.g., python parse_pages_articles_multistream.py ~/wikiann/latest/yowiki-latest/yowiki-latest-pages-articles-multistream.xml.bz2 ~/wikiann/latest/yowiki-latest/yowiki-latest-pages-articles-multistream-index.txt.bz2 ~/wikiann/latest/yowiki-latest/output --nworker 4

Step 3: annotate (apply sentence segmentation, tokenization, etc.) parsed output

 usage: annotate.py [-h] [--nworker NWORKER] [--verbose] indir outdir lang

 positional arguments:
   indir                 input directory (blocks/)
   outdir                output directory
   lang                  wikipedia language code

 optional arguments:
   -h, --help            show this help message and exit
   --nworker NWORKER, -n NWORKER
                         number of processors to use (default=1)

 e.g., python annotate.py ~/wikiann/latest/yowiki-latest/output/blocks ~/wikiann/latest/yowiki-latest/output/blocks.pp yo --nworker 4

Wikidata

Import Wikidata json dump (.json.bz2) into Mongodb and create index

• Index:

  Wikidata ID: { id: 1 }

  English Alias: { aliases.en.value: 1 }

  English Wikipedia Title: { sitelinks.enwiki.title: 1 }

  Freebase ID: { claims.P646.mainsnak.datavalue.value: 1 }

  subclass of: { claims.P279.mainsnak.datavalue.value.id: 1 }

  instance of: { claims.P31.mainsnak.datavalue.value.id: 1 }

  all properties: { properties: 1 }

• Partial Index for Covered Query: { sitelinks.enwiki.title: 1, id: 1 } { labels.en.value: 1, id: 1 }

• Performance: ~3 hours for importing, ~1 hour for indexing (--nworker 12, --chunk_size 10000, based on 20180717 dump (25 GB))

Quickstart

 cd wikiann/wikidata/mongodb/

Step 1: import

usage: import.py [-h] [--chunk_size CHUNK_SIZE] [--nworker NWORKER]
                 inpath host port db_name collection_name

positional arguments:
  inpath                Path to inpath file (xxxxxxxx-all.json.bz2)
  host                  MongoDB host
  port                  MongoDB port
  db_name               Database name
  collection_name       Collection name

optional arguments:
  --chunk_size CHUNK_SIZE, -c CHUNK_SIZE
                        Chunk size (default=10000, RAM usage depends on chunk
                        size)
  --nworker NWORKER, -n NWORKER

Step 2: index

usage: index.py [-h] host port db_name collection_name

positional arguments:
  host             MongoDB host
  port             MongoDB port
  db_name          Database name
  collection_name  Collection name

Miscellaneous

• If you get errno:24 Too many open files error, try to increase system limits. For example, in Linux, you can run ulimit -n 64000 in the console running mongod.