Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-2031
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-20T18:13:49.946118Z
status: published
lens: g_space
sector: labor
headline: Use of Large Language Models (LLM) for Occupational Classification: An example of Early Learning & Child Care (ELCC) Workers
dek: IntroductionThe use of Artificial Intelligence (AI) for occupational classification has drawn the attention of statistical agencies because manual coding is time-consuming and expensive. In Canada, the national "$10-a-day" Early Learning and Child Care (ELCC) program has increased the need for more detailed information on the ELCC workforce. ObjectivesGiven the successful use of Machine Learning (ML) for occupational coding in open-ended Census write-in responses, this study explored the feasibility of creating…
gain_title: Few-shot LLM classification, paired with expert review, split broad occupation groups into finer subgroups that better identify child care workers than manual coding would have allowed efficiently.
problem_title: (none)
trace_subject: (none)
gain_reading: Few-shot LLM classification, paired with expert review, split broad occupation groups into finer subgroups that better identify child care workers than manual coding would have allowed efficiently.
gain_evidence: The use of LLM resulted in a more granular and meaningful occupational classification for ELCC workers.
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers tested whether a large language model could code open-ended occupation answers from Canada's 2021 Census of Population, focusing on the early learning and child care sector. They processed 57,135 write-in responses with Meta AI's Llama 2 using few-shot prompting, in which the model gets only a handful of examples.

The work matters because the national $10-a-day program has raised demand for detailed information about this workforce, and existing categories were too coarse for policymakers. The authors report the tool helped only when combined with subject-matter expertise, so it does not replace expert judgment, and the abstract gives no accuracy figures.
limitation: The authors say the LLM tool cannot offer improvements in isolation and must be complemented by subject-matter expert input.
tag: Evidence-backed gain
key_points: The study used 57,135 write-in responses from Canada's 2021 Census of Population. | Llama 2 was applied with few-shot learning, using a limited number of examples to guide classification. | Existing NOC and NAICS categories were not specific enough to identify a policy-relevant group of child care workers. | Expert input and iterative review of LLM output were used to refine NOC groups into several subgroups.
rundown: Statistical agencies have been interested in AI for occupational classification because manual coding is slow and costly. The study was motivated by earlier successful machine learning use on Census write-in responses and by Canada's growing child care sector.

The authors found that the standard NOC and NAICS classifications could not isolate a meaningful group of child care workers. Through repeated review of LLM classifications guided by subject-matter expertise, they divided NOC groups into subgroups relevant to the sector.

They conclude that manually assigning workers to the refined subgroups would have been inefficient and onerous. They also state that the LLM cannot deliver improvements alone and needs expert input.
sources:
- peer_reviewed | International Journal of Population Data Science | https://doi.org/10.23889/ijpds.v11i1.3367 | 2026-10-19
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

07be0986f81075dbc7c95de134f8465fcee9e2e8ff4f95006f03a2623a1dfb66

previous

0000000000000000000000000000000000000000000000000000000000000000