Hybrid Classification of European Legislation using Sustainable Development Goals

. This study focuses on the automatic classification of European Union legislative documents according to the United Nations Sustainable Development Goals (SDGs) to monitor and improve government policies and legislation. This allows ex-ante checks during legal drafting to better align the new proposal with SDG policies and ex-post evaluation tools for monitoring the implementation and effectiveness of SDG strategies in the European legislation over time. The research aims to assess the alignment of legislative efforts with these global goals by utilizing an extensive corpus of regulations and directives from the Juncker (2014–2019) and von der Leyen Commission periods (2020–2024). The proposed Hybrid AI methodology employs an unsupervised deep learning approach, leveraging the structure of legislative documents formalized in the Akoma Ntoso XML standard. The research has two primary objectives: first, to examine a novel weighted approach where classifications of the initial articles guide the classification of subsequent articles using an unsupervised sentence embedding model. Second, to monitor document-level classifications over time, tracking legislative evolution and comparing policies under different European Commission presidencies. Initial findings, based on legal expert validation of the technical f indings expressed to metrics, reveal that the first articles of legislative documents are crucial in determining the correct SDG classifications and that these classifications may evolve over time with normative modifications and new strategic policies.