Address Data Parser with AI and Nominatim

In this project, we are looking for a developer to create pure java artificial intelligence based console application, using a machine learning library such as Weka ([login to view URL]) in concert with OSM’s Nominatim search library ([login to view URL]) that will parse dirty address data from financial transactions. The purpose of the application is to parse a single line of address data, containing addresses that may be associated with any country into the an output that contains certain basic elements: Name: The name of an entity or individual; Address_Line_1: The first line of a business address, not including a suite or floor number; Address_Line_2: The second line of a business address, that should only include a suite or floor number; City: The city of the address Subdivision: The state, or province or other country subdivision of the address; Postal_Code: The postal or ZIP code of the address; and Country: The country of the address (in 2 character ISO format) Some addresses in the data set should not be considered parsable, and those addresses are only considered addresses that do not contain at least a street address (Address_Line_1) and one other address feature. Success on this project will be deemed to be that the final application will have the ability to properly parse 90% of all parsable addresses in a test data set of 10,000 records. We will provide sample data in two phases: 1. Initial phase – 1000 Addresses (provided only to selected developer); 2. Developer Testing Phase – 10000 Addresses After developer testing, we will perform our own test before work acceptance on a separate set of 10,000 addresses. Only once the benchmark performance established above is met, we will release the final milestone. Example for this parsing might be as follows: Example 1: FAKE COMPANYNAME LLC 6421 LAKE WASHINGTON BLVD NE SUITE 101 KIRKLAND WA 98033 Name: Fake Companyname LLC Address_Line_1: 6421 Lake Washington BLVD NE Address_Line_2: Suire 101 City: Kirkland Subdivision: Washington Postal_Code 98033 Country: US In this case, all data was easily parsed, however the country code of US would need to be automatically appended by the process. The country should be recognized based on the format of the ZIP code and the State code of Washington. Example 2: SOMEADDRESS, Inc. 51 MOORGATE LONDON EC2R 6BH Name: SomeAddress, Inc. Address_Line_1: 52 Mooregate Address_Line_2: Null City: London Subdivision: England Postal_Code Ec2R 6BH Country: UK In this case, most data was easily parsed. The Subdivision and Country were appended, and should have been recognized by the AI process based on recognition of the Postal Code and City. Example 3: RENTEN SERVICE PAYER DEUTSCHE POSTBANK AG KENNEDYALLEE 62 - 70 BONN DE Name: Renten Service Payer Deutsche Postbank AG Address_Line_1: Kennedyallee 62-70 Address_Line_2: Null City: Bonn Subdivision: North Rhine-Westphalia Postal_Code: 53175 Country: DE In this case, the German address required the append of the postal code, and subdivision, which was obtained from Nominiatim after the address was parsed. All other data was present. Example 4: FRANCOIS HOLLANDE 35 WOODSFORD SQUARE LONDON Name: Francois Hollande Address_Line_1: 35 Woodsford Square Address_Line_2: Null City: London Subdivision: England Postal_Code: W11 4PY Country: UK Searching the address string provided, 35 Woodsford Square London against Nominatim finds that this address only exists in London UK enabling population of the Subdivision, Postal Code, and Country.