Big data analysis of graduate industry talent demand using deep pretrained language models

Traditional labor market studies have primarily focused on the supply side, which limits understanding of graduate talent demand across industries. In the context of digital and green transitions, this study develops a deep pretrained language model framework to analyze graduate talent demand using a large volume of recruitment data. The analysis draws on 1,048,365 recruitment postings from China. The framework combines a refined BERT-Base-Chinese model with Chinese text mining techniques. Distinct training datasets were constructed for the three classification tasks, achieving accuracies of 73.93% for industry classification, 83.59% for matching between academic majors and job positions, and 78.15% for job level classification. The findings show that industrial upgrading and labor market segmentation have occurred concurrently. Industries with strong technological intensity, such as manufacturing, information transmission, education, and scientific research, show stronger requirements for professional alignment and sustained wage growth. In contrast, service industries, such as accommodation, catering, wholesale, and retail, show broader job structures, a higher concentration of junior positions, and slower wage growth. Mining and quarrying, scientific research and technical services, and electricity, heat, gas, and water supply show the highest levels of professional alignment and entry barriers, reflecting a stronger reliance on specialized qualifications. Junior positions in service industries show greater fluctuation and lower specialization. These results indicate that professional alignment remains an important determinant of job quality and wage progression. They also highlight the importance of human capital development across industries under digital and green transitions. The findings further suggest the need to strengthen talent cultivation aligned with industry demand and to establish dynamic labor market monitoring systems, so that curricula and qualification structures can better respond to evolving industry competency demands.