MEET-LM: A method for embeddings evaluation for taxonomic data in the labour market

Lorenzo Malandri, Fabio Mercorio, Mario Mezzanzanica, Navid Nobani

Abstract: Highlights•we define and formalise a method – namely MEET-LM – to evaluate and select embeddings from a large text corpus that preserves the co-hyponyms relationships synthesised from a domain-specific taxonomy.•we define a novel metric to evaluate hierarchical semantic relatedness between concepts of taxonomy to evaluate embeddings trained on a text corpus of ICT-related online job vacancies.•we implement, apply and evaluate (extrinsic and intrinsic) MEET-LM to 2+ million real ICT-related vacancies collected in 2018 in the UK, as research activity of an EU project.•We show MEET-LM effectively encodes the co-hyponyms relationships synthesised from the standard ICT occupation taxonomy.