|Perl: the Markov chain saw|
Re: Re: Beast of the Number: Parsing the Feral Phoneby mojotoad (Monsignor)
|on Apr 17, 2002 at 22:25 UTC||Need Help??|
Interesting data regarding Germany -- I had no idea.
It was for precisely this sort of reason, however, that I made no attempt to try and figure out area codes for numbers from various countries around the world. The result of this parsing could easily be passed along to a country-specific module for more appropriate parsing and beautification.
There are a couple of things to point out here that I did not mention in the article (due to 64k limit on nodes). I made no attempt to parse valid IDD prefixes even though lists for each country are available on the net. The reason is that the IDD prefixes are not mutually exclusive to the Country Codes. Nor, unfortunately or unexpectedly, are area codes for a particular locale.
This reality produces ambiguous areas where I could be slurping up an area code or IDD as a country code. What's needed in *that* case is some concept of the natural phone number length for that locality. Rather than get that specific, though, I relied on a threshold length and size percentages as measured against the remainder of the number. It's not perfect, but for my data set it worked suprisingly well.
Given your information about the variability of German numbers, particularly the 14-digit monsters, this technique might fail if the area/province codes happen to match valid country codes elsewhere. This of course only applies to numbers that are presented *without* their Country Code.
Once this code has its hands on what it thinks is the local number, it's just stored as a single number. I chunk it for display purposes, but generically and in a U.S.-centric kind of way: 4 digits on the suffix, preceded by groups of three digits as long as there are digits left.
Also, keep in mind that this code is intended to operate on raw, unrestricted data fields. Typoes, blippoes (???? or xxxx) and all of it are present in the data. There's not a whole lot you can do in these cases to pull out a valid number without knowing in excruciating detail the particulars of the intended country.
GIGO, GIGO, it's off to the dumpster we go!
BTW, I suspect this code might take quite a while to run on 10 million numbers, even if they are well-behaved.
Thanks for the comments and feedback. Any thoughts on whether any of this should be CPAN-bound once cleaned up? (new names and POD, obviously, but beyond that...)