Street address locations extracted ...

Software views

Street address locations extracted from unstructured html Tue, May 15. 2007

Many sites have their street address location in an unstructured html page which has a wide range of uses. From making searches for your local computer store easier to plotting hiking trails. The currently problem is getting this info into a structured format which can be geocoded.

MetaCarta has an api to retrieve some of the information, though not street level accuracy.
http://labs.metacarta.com/

True local provides data services
http://www.truelocal.ca

And digital reasoning have a GeoLocator product for retrieving data from unstructured documents.
http://www.digitalreasoning.com/index.php?option=com_content&task=view&id=3

These are fine for large commercial systems, though they are not of much for every day consumer use in tasks like adding a contact to your address book. This task would be seam less if there was an open source project developing this software.

The project could be based on gpl or some other open source license making it free for commercial use. After this key piece of technology is available to every one things like address books, map creating and searching can become seamless.
Posted by Charles Verge in Business Comments: (4) Trackbacks: (0)

Trackbacks
Trackback specific URI for this entry

No Trackbacks

Comments
Display comments as (Linear | Threaded)

#1 - Christopher Schmidt said:
2007-05-15 17:18 - (Reply)

It's not about the code -- the code is simple. (Geo::Coder::US on CPAN does it, and is open source.) Not even the NLP is that hard (Lingua::Tagger::EN will pull out proper nouns to tag.) Instead, the hard part is the data -- many people make millions of dollars a year selling street level accuracy, and they're not about to give it up.

Without the data, you can't do anything useful.

#1.1 - Charles Verge said:
2007-05-22 07:09 - (Reply)

How about a way of doing things that does not require data ? If you look at visually vs as a list of words you might beable to break out the blocks of text which have addresses. From the blocks you could use a text categorization algorithm like Bayesian. After a block is identified something like the geocoder.us or yahoo api can sort out the address into lat and long.

#2 - Christopher Schmidt 2007-05-22 16:33 - (Reply)

I'm not sure what you mean by 'does not require data'.

Bayesian filters require large amounts of data, though not large amounts of geo-data, so it's true that this can be obtained from other sources.

Geocoder.us and the Yahoo API both are data -- they're just APIs to data -- and Yahoo/Google's Geocoders aren't Free.

If you're saying it's possible to take text, make a guess as to what looks like an address, and then use a non-Free API to figure out where that is as a lat/lon -- sure. That's possible. But what's the value in that? Why not just use the MetaCarta API at that point?

#2.1 - Charles Verge said:
2007-05-23 06:38 - (Reply)

Yes, it does require data, though data that would need to be entered on case by case. That is the value, the ability to tune to a specific geographic area.

Here is an example, http://www.hfxnews.ca/index.cfm?pid=1354
When you run that threw the MetaCarta API you get the city Halifax, no street address even though it is clearly a street address. The format of the page is less complex then the MetaCarta contact page, http://metacarta.com/company/contact-us.html

With a system that can be trained this format of address would be picked up after a few corrections. These corrections could only be done by a person who knows the way people enter addresses in that geographic area. This is because people the local standard rather then the international one.


Add Comment

Enclosing asterisks marks text as bold (*word*), underscore are made via _word_.
Standard emoticons like :-) and ;-) are converted to images.

To prevent automated Bots from commentspamming, please enter the string you see in the image below in the appropriate input box. Your comment will only be submitted if the strings match. Please ensure that your browser supports and accepts cookies, or your comment cannot be verified correctly.
CAPTCHA