Our paper, \u201cOn the Similarity of Occupations,\u201d has been accepted for the Researching Work & Learning International Conference (RWL) 2026 in Brisbane!<\/p>\n\n\n\n
This paper examines how the similarity of occupations can be measured. To this end, it presents and combines two complementary perspectives: lexical similarity<\/em> captures differences in the spelling of job titles, while semantic similarity<\/em> takes into account how similar occupations are in terms of their content.<\/p>\n\n\n\n To measure semantic similarity, job titles are represented as vectors; the cosine distance\u2014that is, the angle between two vectors\u2014is then calculated. The job titles \u201cprogrammer\u201d and \u201csoftware developer\u201d are semantically similar. The corresponding vectors point in a similar direction, while the vector for \u201cmedical doctor\u201d points in a different direction.<\/p>\n\n\n\n The evaluation based on the ISCO and KldB taxonomies shows that both measures capture different aspects of occupational similarity and complement each other effectively. In particular, semantic similarity largely reflects hierarchical relationships within occupational classifications. At the same time, several anomalies within the KldB were uncovered.<\/p>\n\n\n\n This approach lays the groundwork for numerous further steps, such as the automatic extraction, classification, and semantic interpretation of occupations.<\/p>\n\n\n\n