Placename matching algorithms and services for geocoding

Question

Are there any placename-specific matching algorithms, routines or services available? By placename specific I mean not just a simple string-difference match, or soundex/metaphone, but something that understands geographic abbreviations and conventions.

For example, match "W. Mt Washington Pkwy" with "West Mount Washington Parkway"; or "Oxon" with "Oxfordshire"; or "Oxford-st" with "Oxford Street"; or "N.S.W." with "New South Wales"; or "1234 Main St NW" with "1234 Apartment 345, North West Main Street", or "St. John's Ferry" with "Saint John Ferry".

Most of the addresses I'm dealing with (to geocode, to plot on maps for historical GIS) are at the town level. In most cases there are further elements of the address available (country, state, province, country) to avoid too many false positives. I do have data to match against, the challenge is to get the names to match, allowing for varied spellings and abbreviations.

There are several stages to match, assuming we're in the same state/province, including but not limited to:

• simple string matches, allowing for differing word order
• expand basic geographic abbreviations (W, St, Cty, and so on), based on the language(s) and conventions of the country, really just a bunch of look-up tables
• expand local admin area abbreviations and codes (NY, NSW, Oxon, and so on)
• look up placename changes and language alternatives

This is all basic stuff (the later three depend more on data than on an algorithm). Although some parts are tricky, like working out when to expand St as Saint, or Street (or try both).

So I assume this widely needed problem has been solved many times before (by people much more expert than me!), so I shouldn't re-invent the wheel.

But, even limiting it to English language countries (UK, Canada and so on, as well as the US), I haven't been able to find any published work or software for this. I must be looking in the wrong places. Does anybody have any suggestions for existing work I can build on?

Answer 1

ESRI has a tool for ArcMap, called Standardize Addresses which breaks up an address into parts (which appears to be based on somewhat on URISA's standards for addressing). From there, the fields can be concatenated and geocoded. If you didn't want the abbreviations, it's a lot easier to do a massive find and replace as the name is broken up into separate cells. I learned too quickly that replacing ave with avenue on the full street name will give annoying results such as Shaver -> Shavenuer. While that obviously won't help with all cases, it does account for a few.

It shouldn't be too tricky to determine when to expand St as Saint vs Street. A rough way to do it would be to split the string by spaces and then search the list of substrings for occurrences of the common post types (dr,ln,blvd,etc). If only one ST is found, then it generally is Street. If there is an ST found with another post type, then it generally is Saint.

Answer 2

Mapzen has a pretty impressive international address parser, libpostal. https://mapzen.com/blog/inside-libpostal/

Have spent a lot of time searching for the same thing, and this is the best I've found so far.