In this paper, we analyzed a large set of online job postings from Latin America, using natural language processing techniques to create a new, detailed classification of job skills. Specifically, we compared two methods used to evaluate the similarity between occupations: a “revealed comparative advantage” (RCA) approach and a modern embeddings-based approach using OpenAI's models.

We found that the new approach, which uses modern language tools, can provide a more accurate and up-to-date picture of the skills needed for different jobs. This can help governments and organizations better understand job markets and make more informed decisions.