LLM Classifies Canadian Child Care Workforce

Researchers Maria Ahmed, Kristyn Frank, Anne Guèvremont and colleagues reported a study in the International Journal of Population Data Science in 2026 on using a large language model (LLM) to classify Early Learning & Child Care (ELCC) workers in Canada. They examined whether an in-context classification process could provide more detailed occupational information for the sector, amid rising demand associated with Canada’s national “$10-a-day” ELCC program.

The team applied Meta AI’s Llama 2 to 57,135 open-ended job responses from Canada’s 2021 Census of Population. The model used few-shot learning, in which it receives a limited number of examples, to assign occupational classifications. Researchers found that existing National Occupational Classification (NOC) and North American Industrial Classification System (NAICS) categories were not specific enough to identify an ELCC workforce group useful to policymakers.

Subject-matter experts and iterative reviews of the LLM’s classifications refined NOC occupational groups into subgroups covering jobs relevant to the ELCC sector. The researchers said assigning workers to these more detailed groups through typical manual methods would be inefficient and onerous. They concluded that LLM classification produced a more granular and meaningful classification efficiently, but said it needs input from subject-matter experts and cannot improve classifications on its own.

Researchers Maria Ahmed, Kristyn Frank, Anne Guèvremont and colleagues reported a study in the International Journal of Population Data Science in 2026 on using a large language model (LLM) to classify Early Learning & Child Care (ELCC) workers in Canada. They examined whether an in-context classification process could provide more detailed occupational information for the sector, amid rising demand associated with Canada’s national “$10-a-day” ELCC program.

The team applied Meta AI’s Llama 2 to 57,135 open-ended job responses from Canada’s 2021 Census of Population. The model used few-shot learning, in which it receives a limited number of examples, to assign occupational classifications. Researchers found that existing National Occupational Classification (NOC) and North American Industrial Classification System (NAICS) categories were not specific enough to identify an ELCC workforce group useful to policymakers.

Subject-matter experts and iterative reviews of the LLM’s classifications refined NOC occupational groups into subgroups covering jobs relevant to the ELCC sector. The researchers said assigning workers to these more detailed groups through typical manual methods would be inefficient and onerous. They concluded that LLM classification produced a more granular and meaningful classification efficiently, but said it needs input from subject-matter experts and cannot improve classifications on its own.