Learning Skill Representations from Recruitment Data for Skill Gap Analysis and Career Path Optimization

Abstract

To address the mismatch between educational provision and labor-market demand affecting the employment of arts graduates, this paper proposes a skill-gap analysis framework integrating recruitment-text mining, semantic representation learning, and knowledge graphs. First, a corpus of job advertisements is compiled from online recruitment data, and a Transformer-based model is used to encode job descriptions semantically, enabling the extraction of skill entities and the development of context-sensitive representations. Second, semantic skill embeddings, hierarchical clustering, and BERTopic-based topic modeling are combined to construct a skill-classification system for arts-related positions, with the aim of identifying core skill groups and emerging competency requirements in the employment market. Furthermore, the proposed framework maps job positions, skill requirements, and students’ existing abilities into a shared semantic space and uses knowledge graphs to characterize structural relationships among jobs, skills, and competencies. On this basis, distributions representing market demand and educational skill supply are constructed and compared to quantify gaps affecting arts graduates. The reported experimental findings suggest that requirements for arts-related positions increasingly combine conventional professional expertise with digital technology, AI-assisted creative practices, data literacy, and transferable professional competencies. The results also identify apparent shortcomings in university preparation concerning computing skills, problem-solving, project implementation, and adaptability across disciplinary boundaries. Together, these analyses offer a structured basis for interpreting skill mismatches and informing curriculum development, individualized upskilling, and career planning.