Need an efficient way of address comparisons

Question

I am trying to find out a way to compare 2 addresses
e.g. 6841 Day Drive with 6841 Day Dr.. In this example Drive and Dr needs to be matched somehow
I read some forums which say split the text and then try to match the numerical portion and then street name.
Now imagine there are thousands of records to be processed. Definitely the above process is practically useless when processing so many records
Can anyone help me with giving some ideas how to implement that efficiently.
Thanks in advance,
Mandeep

Answer 1

JoachimSauer wrote:
mandy_m wrote:
I call it practically useless to split the string and do text comparisons considering the fact that there are many ways in which a user might write the address.
I gave an example of Drive - Dr, others may be Road - Rd, Street - St, Aprtment -Apt..... and the list is very longThat's why I said you need to canonicalize the data. Write a method (or class) that canonicalizes any given address. Then apply that to all data that you store (i.e. all your existing data). This might take some time, but only needs to be done once.Joachim's defnitely giving you good advice. If you have no control over the data coming in, this is definitely not a simple job; if you do, you could require people to put the 'type' of street and the building number separately from the street name (and don't forget that building numbers can be ranges if you're dealing with corporate addresses).
In England, a lot of people give names to their houses, which further muddies the waters as it is often the first line in the address (the same could be true of a corporate address: Company House, 200-250 Suchandsuch Rd....etc).
A few things to try (some have already been suggested):
1. Create an Address class based on the components that you know you want. At the very least, it should have 'equals()', 'hashCode()' and 'toString()' implemented.
2. Convert the input address to lowercase (or uppercase).
3. Create a dictionary of known street types (road, avenue, boulevard...etc). This could be a static HashSet in your Address class.
4. Create a list of known abbreviations (apt, rd, dr...). I'd suggest you include versions with and without punctuation.
5. Expand all abbreviations in your input address.
6. Remove all remaining punctuation, but keep the original line orientation (this might involve splitting your address string into 'lines' based on commas for example).
7. Look for a line that contains one of your known 'street types'. This is highly likely to be the actual street address.
8. Extract the number (or numbers, in the case of a range). Watch out for things like '221b baker street'.
9. Load the resulting components into your Address variables.
The fact of the matter is that even with all of the above, you will probably not have a 100% solution. Personally, I'd reject all input addresses that are so badly written that you can't determine required components for your class, but I'm an ornery cuss...
Apart from that, you could check out the postal service website where you live to see if they have any guidelines for parsing addresses. The GPO here used to have a great page for that.
Good luck. What you want is not simple.
Winston