I discovered the data is available up to date. Maybe soon or later I'll repeat and extend the analysis, potentially also using multiple ways to compute the vectors, including SBERT (or better SModernBERT).