A Transformer-Based Intelligent System for Hierarchical Occupational Classification in the Labor Market

The classification of job postings into standardized occupational categories is a challenging task due to the unstructured, heterogeneous, and noisy nature of labor market data. This process is particularly relevant for labor market analysis conducted by government agencies and employment services, which require accurate and consistent occupational classifications to support public policy, workforce development, and training investment decisions. Manual classification is time-consuming and prone to inconsistencies, highlighting the need for scalable, intelligent systems. This study presents an applied artificial intelligence framework that integrates unstructured textual data from multiple online job boards with structured occupational taxonomies. The dataset comprises 4,605 manually labeled job postings covering 104 occupational classes, ensuring balanced representation across sectors and levels of specialization. We fine-tuned BETO, a Spanish-language variant of BERT (Bidirectional Encoder Representations from Transformers), to perform large-scale hierarchical classification of job descriptions mapped to the <italic>Clasificaci&#x00F3;n Internacional Uniforme de Ocupaciones</italic> (CIUO 08.CL), the Chilean adaptation of ISCO-08 (International Standard Classification of Occupations). From an engineering standpoint, the model is integrated into a national labor market analytics platform to support real-time occupational demand monitoring and institutional decision-making. Our system outperforms traditional machine learning and deep learning baselines, achieving a weighted accuracy of 0.69 and an F1-score of 0.66 at the four-digit classification level. The proposed transformer-based architecture offers a robust and scalable solution for real-world labor market intelligence applications.