Whole of Australian Government Occupation Coding Service

In 2023, the ABS was funded to deliver a new occupation data coding service for the whole of Australian government. In this context, ‘coding’ means matching free text to numbered and labelled categories that are meaningful and comparable across data sources. 

The coding service is currently being developed by the ABS and tested with a range of key agencies. 

This exciting new capability will make classifying data to the latest official standard classifications easier, faster and cheaper. It will use cloud compute for power and scale to achieve highly automated, high-quality coding. 

The service will enable data coding to be consistent across the Australian data landscape. This means data assets will be more coherent for research and analysis — which results in better information for policy makers! 

The coding service will classify single data records in real time, supporting online surveys and agency website forms. It will also quickly and efficiently classify large datasets in batches, up to millions of records.

When the service goes live in June 2025, it will classify data to a choice of the:

• Australian and New Zealand Standard Classification of Occupation (ANZSCO) 2022 Australian Update
• Occupation Standard Classification for Australia (OSCA) 2024, Version 1.0. 

The service will be maintained by the ABS to ensure the most up-to-date occupation classifications are available.

The coding service will use models trained by machine learning (ML) to code data. ML is a branch of artificial intelligence (AI), that uses algorithms to analyse, draw conclusions and make predictions from patterns in data. The algorithms are trained to spot patterns in the training data, which helps improve coding accuracy. ML mimics the way that humans learn, gradually improving in accuracy as models are trained and retrained.

An advantage of the machine learning models in the coding service is they are being trained on ABS data with text that represents the Australian population. This means the coding service will be able to interpret and classify occupations described in the common language and broad vocabulary used by Australians.

For example, when a person types ‘electrician’ into an online form, the coder will recognise that this belongs to ‘Electrician (General)’. With the benefit of ML training, the coder will be able to recognise and classify a range of text responses, including misspellings and colloquial terms like ‘sparky’.

The coding service models are being developed within ABS infrastructure. This development follows strong privacy and security practices within existing ABS safeguards, while ensuring responsible and transparent use of ML technology. 

The service will use a highly controlled, narrow ML capability. The models will be purpose-built, trained on ABS data, and designed to only classify free text to official statistical classifications. The coding service will not be able to generate responses, provide insights, or perform any function beyond this specific task. The models will remain static until being manually updated or retrained in a controlled process. The models will not learn from the data that users send to be coded.