Occupation classification model based on DistilKoBERT: using the 5th and 6th Korean Working Condition Surveys.

Area of Science:

• Computational linguistics
• Occupational health
• Data science

Background:

• Accurate occupation classification is crucial for policy development and epidemiological research.
• Existing methods may not fully capture the nuances of job responsibilities.
• Automated systems can enhance efficiency and accuracy in this classification process.

Purpose of the Study:

• To develop and evaluate an occupation classification model using DistilKoBERT.
• To assess the model's performance using key evaluation metrics.
• To explore the potential of AI in occupational health and safety research.

Main Methods:

• Utilized data from the 5th and 6th Korean Working Conditions Surveys (2017, 2020).
• Employed natural language responses of 99,665 Korean workers and Korean Standard Classification of Occupations (3-digit codes).
• Fine-tuned a DistilKoBERT model on a 7:3 training-test split dataset.

Main Results:

• The DistilKoBERT model achieved an overall accuracy of 84.44% in classifying 28,996 participants into 142 occupational codes.
• Weighted precision, recall, and F1 scores were 0.83, 0.84, and 0.83, respectively.
• The model showed high precision for service/sales workers and occupations prevalent in the training data, with lower precision for managers.

Conclusions:

• An automated occupation classification system based on DistilKoBERT demonstrates promising performance.
• The model offers a valuable tool for epidemiological studies in occupational safety and health.
• Further enhancements are needed to improve classification accuracy across all occupational groups.