Guidelines to Improve Prediction Accuracy

In order to get the most accurate prediction of address components, your input address strings should adhere to these patterns.

Guidelines for United Kingdom Addresses

Avoid non-address components
Presence of non-address components in the input string might lead to wrong prediction. Remove such components before feeding the string for prediction.
Maintain a sequence in address components
The address components should be placed in this order: OrganizationName > Floor > PlaceName > AddressNumber > Street > Neighbourhood > City/Suburb/County > PostCode > Country.
Example:
  • Pitney Bowes Limited London Milenium street Unit 3 AB10 3DF GBR
  • Pitney Bowes Limited Unit 3 Milenium street London AB10 3DF GBR
Remove redundant address components
The input address string should not have repeated address components, such as two different organization names or repetitive name of an organization in one string.
Example:
  • Pitney Bowes Limited Pitney Bowes Limited Unit 10 Logix Cyber Park 10 Manor Street London AB10 3DF GBR
Follow single-token organization names with organization type
A single-token organization name should be followed by the type of the organization, such as Ltd, Inc, and Reg. In the example below, Ardian is a single-token organization name. In this case, the organization name is not followed by the type "Limited," and the results may be inaccurate.
Example:
  • Ardian Fourth Floor Channel House St Helier Je2 4UH GBR
  • Ardian Limited Fourth Floor Channel House St Helier Je2 4UH GBR

Limitations in United Kingdom Addresses

An address string of any of these kind is susceptible to getting inaccurately predicted by the address parser. Watch out for these in your address strings.

Presence of another address component as name of the organization
If the name of the organization includes any other address component, such as Floor, Flat, and House, the prediction accuracy may be affected.
Example: Flat Seasons 632 Kings Road London Middlesex SW6 2DU GBR
Organization name having numbers
If an organization name has numbers, it is susceptible to getting erroneously predicted.
Example: 123 Limited ABC Street AB10 3DF GBR

Guidelines for German Addresses

Avoid non-address components
Presence of non-address components in the input string might lead to wrong prediction. Remove such components before feeding the string for prediction.
Maintain a sequence in address components
The address components should be placed in this order: OrganizationName > Floor > PlaceName > AddressNumber > Street > Neighbourhood > City/Suburb/County > PostCode > Country.
Example:
  • 3 Weseler Strasse 46514 Schermbeck DEU
  • Weseler Strasse 3 46514 Schermbeck DEU
Remove redundant address components
The input address string should not have repeated address components, such as two different organization names or repetitive name of an organization in one string.
Example: Weseler Strasse 3 Weseler Strasse 46514 Schermbeck DEU
Ensure address number and street name are included
Your address string needs to have address number and street name. Missing out these essential address components will impact the accuracy of the result.
Example:
  • 46514 Schermbeck DEU
  • Weseler Strasse 3 46514 Schermbeck DEU
Do not have merged components in address strings
Merged address components result in incorrect prediction.
Example:
  • Weseler-Strasse-3 46514 Schermbeck DEU
  • Weseler Strasse 3 46514 Schermbeck DEU
Avoid addressee name in the string
Address name in the string results in incorrect prediction for the German addresses.
Example:
  • Mr John Doe Weseler Strasse 3 46514 Schermbeck DEU
  • Weseler Strasse 3 46514 Schermbeck DEU
Do not have bracketed "()" address component
Including any of your address components inside brackets "()" will leave it unparsed.
Example: Weseler Strasse 3 46514 (Schermbeck) DEU