The Anomebot2

The bot is back running again. I've tried to minimize disruption, but there is a good chance some systematic error patterns will return, since code changes made over more than a year have been lost. I'm going back over old comments, and see if I can find the comments and reinstate these fixes. Until then, please be forgiving; I'm making every effort to bring things back to where they were before. — The Anome (talk) 13:01, 24 November 2024 (UTC)

Emergency bot shutoff button

Administrators: Use this button if the bot is malfunctioning. (direct link)
Non-administrators can a malfunctioning bot to Wikipedia:Administrators' noticeboard/Incidents.

Note: Blocking will stop further edits: the bot will intermittently retry errors for several minutes, but should then automatically shut itself down until restarted manually; please use a ten minute block or longer to be sure of stopping it.

This bot is designed to add standardized machine-readable geodata records to relevant articles in the English-language Wikipedia, using data from GNS, GNIS, OSGB coordinates in UK articles, plaintext geodata scraped from article text, and interwiki-linked geotag data from other-language Wikipedias. -- The Anome 12:13, 22 September 2007 (UTC)

Status

• 125,000+ geotags added to date, based on data from a variety of sources

• 102,000+ more articles identified as potentially eligible for tagging have been marked using {{coord missing}}

• 61,000+ articles reformatted from old geotag formats to use {{coord}}

- The Anome (talk) 14:10, 13 September 2008 (UTC)

Update: As of 2009-06-12:

• 365,800 distinct articles edited -- The Anome (talk) 17:08, 12 June 2009 (UTC)

Recent activities

Currently backfilling a number of corner cases missed by earlier over-cautious heuristics, using:

• machine parsing of plaintext geodata found in dumps
• automatched GNS data
• interwiki-matched machine-readable geodata from other language editions

This is very laborious for the bot, as it requires the re-scanning of large numbers of false positives, and will result in only a few hundred articles being geocoded, but machine time is cheap, the re-scans are necessary in any case, and this will lay the foundations for larger systematic efforts to come later.

-- The Anome (talk) 14:30, 14 September 2008 (UTC)

Done. -- The Anome (talk) 04:29, 15 September 2008 (UTC)

Current activity

Finishing adding a large number of {{coord missing}} tags. Almost complete. -- The Anome (talk) 23:50, 13 October 2008 (UTC)

To do

Geotags:

• Standardize existing geotags. "coor title *" is now done, "coor *" pending.
• Finish adding "coord missing" to all eligible articles.
  • Ancient sites should be templatable as missing, while still blocked from being given coordinates automatically.
  • Go back and use CatScan to find any remaining franchises mis-tagged as "coord missing".
• Rebuild article state map from log file and other stored data.

Interwiki:

• In the absence of up-to-date Kolossus data, start using the externallink table API to live-scan non-en: Wikipedia editions for URLs in order to obtain interwiki patterns
• Use full interwiki data to regenerate fuller tags where only KML data was used for earlier tagging.

Consistency and correctness:

• Use 1-degree-tile binning to look for outliers
• Look for misuse of coord tags for offplanet locations: report to WikiProject for fixing

Matching:

• Hierarchical matching with disambiguation by subnational entities; rejected some time ago because ineffective, but may have become possible with greater navbox systematization in last year
• Open research topic: Bayesian inference of relative locality from the link graph -- this may be an effective way of handling the above. Use places with known locations as training set.
• Properly handle undersea features and disputed territories with no applicable recognized country
• Types of places not yet keyword-matched during graph traversal:
  • Casinos
  • Resorts
  • Historic districts [?]
  • Ports and harbo[u]rs by country
  • Bus and some metro stations

New data sources:

• Collect lists of country-specific coordinate data
• Mine geodata from images included in articles (thanks to User:Planemad for the suggestion)

Infoboxes:

• Scan for unusual/broken parameters in infoboxes.
• Start work on standardizing infoboxes.

-- The Anome (talk) 13:47, 12 October 2008 (UTC)

Forthcoming attractions

With >70,000 data points, I now have enough data to do a spatial analysis of the category tree, and to generate lists of possibly misclassified or mislocated outliers. The cleaned up bounding data could then be used as a Bayesian classifier for future work. -- The Anome 10:14, 24 August 2007 (UTC)

• The category+link graph may be a better choice for this. -- The Anome (talk) 13:59, 12 October 2008 (UTC)

Ambiguity problems

Because of severe name ambiguity problems,

• Japanese locations are now filtered out of most machine-matched geodata sets.
• Recent Canadian data has had similar problems, and is now also filtered from the output of several matching algorithms.
• Because of numerous bizarrely-formatted disambig pages which confused the matching algorithms, Polish locations are also filtered.

-- The Anome (talk) 13:58, 12 October 2008 (UTC)