Introduction
This tool is running live using the Wikidata SPARQL query service and the OSM Overpass and Nominatim APIs. It works best with a city, island or administrative area.
The index page displays list of existing matches and a search box. If you pick an existing result you'll see Wikidata items within the given area, for each item there is a list of candidate matching OSM items.
Matching process interface
The matching process takes a few minutes. It downloads potential OSM objects from Overpass and loads them into PostgreSQL with osm2pgsql. The interface provides status updates while the matching process is running.
A large area will trigger an Overpass timeout and the match will fail.
Once the matching process is complete a 'view match candidates' link will appear. The matching is based on names and English Wikipedia categories.
One-to-one between OSM and Wikidata
With this system the aim is for a single OSM entity to link to one Wikidata item. Many geographical entities are represented by multiple objects in OSM with the same name, for example dual carriageway bridges are mapped as two roadways, or buildings within a large site like a hospital.
For bridges with two roadways the system will look for the outline of the bridge tagged with man_made=bridge and for a hospital or other campus the aim is to tag the way or relation that represents the entire site.
English language Wikipedia categories and Wikidata are used for matching
The matching system makes use of categories on Wikipedia because the information on Wikidata is incomplete. Wikidata includes an import of all Wikipedia including the coordinates, but for a lot of items the 'instance of' property is not set.
There is a mapping from Wikipedia category to possible OSM tags. For example if a Wikipedia article has the word 'station' in any category then railway=station is added to the list of possible tags.
Name matching
Each Wikidata name is compared with each name in OSM. Wikidata names are pulled from the item name, alias and sitelinks in every language. Most name fields on the OSM side are considered, old_name and a few others are ignored.
The name comparison includes some normalisation, punctuation is removed.
Existing OSM tags for Wikidata and Wikipedia
Existing Wikipedia tags are ignored. If an OSM entity already has a Wikidata tag it will be left alone.
Items that are unlikely to match
- Radio stations: not mapped on OpenStreetMap
- Streets: the matcher ignores streets to minimise the Overpass data download
- Sites of Special Scientific Interest: not tagged as such on OSM
- Rivers: river relations are not multipolygons so osm2pgsql ignores them
- Things that no longer exist: Wikidata items that existed in the past
- Organisations: Small organisations within office buildings are not in OSM
- Events and festivals: Not part of OSM
Development
- Code: https://github.com/EdwardBetts/osm-wikidata
- Tasks: https://github.com/EdwardBetts/osm-wikidata/issues
- All timestamps stored in the database should be in UTC.