This study presents a framework for automatically constructing skill networks from work history data to enablethe quantitative analysis of engineers’ technical skills. The proposed method consists of three stages: (1) extractingskills from free-text descriptions using a large language model (LLM), (2) normalizing skill expressions using embeddingsand dictionaries, and (3) building skill co-occurrence networks that capture structural relationships amongskills. Applied to work history data across five technical domains in Japanese engineering, the method identifieddomain-specific core skills and quantified their temporal stability over a ten-year span. The analysis revealed thateach domain exhibits a densely connected core structure, with specific technologies and tools consistently occupyingcentral positions. Furthermore, highly connected skills tended to remain stable across years, reflecting the persistenceof core technical competencies in each field. This framework provides a data-driven basis for analyzing engineeringexpertise and supports large-scale skill profiling grounded in empirical engineering work records.
Chujyo et al. (Sat,) studied this question.