Development Seed Blog

The map looked wrong, and when I inspected it closer I saw that it was wrong,
very wrong. Members from Canada were showing up in New Orleans, priests living
in the Vatican appeared in Florida, and the World Bank’s staff was placed in
Washington state instead of Washington, DC.

This is what happened after I tried to map a couple thousand people
worldwide using the Drupal member_geocoder module we’re working on and
will be releasing soon. With at least one member in nearly every country,
the content I was mapping tested the limits of Google and Yahoo’s geocoding
capabilities and showed its limitations. Luckily, I found a solution with the
open source project Geonames, which
several people directed me to after my last post on geocoding.

As I mentioned in my
last blog post, we built a module (member_geocoder module) to improve the mapping
of international content on Drupal websites. The module first looks to Google for
geotagging. If no latitude and longitude points are returned, it switches to Yahoo.
This works great for places like China and England because while Google does not
geocode these countries for legal restrictions, Yahoo does. But unfortunately, Yahoo’s
geocoder doesn’t work that well for other places.

So yesterday I went back to work on the member_geocoder module. Now it only uses
Yahoo to map content in China and England, and still uses Google as the primary
geocoder. The key improvement is that any content that Google can’t geocode and
isn’t in China or England is sent back to Google to with just the country name.
This at least plots content in the correct country. And it gets around the issue
that sometimes Google can’t geocode content because part of the address (like the
street address) isn’t accurate, the content is in a different language, or simply
because Google can’t do it for no good reason.

The real promising work, though, was with Geonames. I found that Geonames was
actually able to geocode at least two records that Google and Yahoo could not.
One was "PRISHTINË, Serbia And Montenegro." If you go to
this
query, you can see that there are multiple spellings and languages stored for PRISHTINË, which
gives Geonames a clear advantage. Conversely, Google only worked with Priština. When you're
dealing with user data, standardizing it is a lot of work that typically doesn’t match up to
where the budget is allocated, so robust solutions that can handle some bumps go a long way.

I’m working now to integrate Geonames into the module, which will hopefully
help expand its accuracy geocoding international points. There is already an
effort to get Geonames into the location module. I will go back and refactor
how the three geocoders fail over to each other in order to contribute back
helpful code and approaches.

See the set of geocoding images here on Flickr