The circular economy (CE) paradigm has recently gained increasing attention in both academic and policy circles. Existing literature has stressed that the transition to the CE paradigm implies innovation aiming to change consumption and production behaviors and technologies. 

Empirical studies have focused on the drivers and effects of the adoption and generation of CE innovations, based on survey and patent data, respectively. However, identifying and tracking CE innovations through patents has been challenging due to the lack of a domainspecific classification system. Existing methods are often insufficient to capture the diversity and complexity of CE technologies. 

This paper proposes a novel methodology for the identification and classification of CE-related patents, combining large language models (LLMs), pre-trained language models (PLMs), and topic modelling techniques. By applying these methodologies to patent data, we uncover significant trends in the distribution of CE patents in sectors, technological fields, and geographical regions. 

Our exploratory findings highlight a growing cross-sector engagement with CE principles, underscoring the transformative potential of circular economy innovations in driving sustainable industrial practices. This paper contributes to advancing the classification and understanding of CE innovations, offering valuable insights to policymakers, researchers, and industry stakeholders. 

Keywords: Circular Economy; Patents; BERT; Large-Language Models. 

JEL Classification: O33. 

Authors:

• Maria Manera 
  University of Torino 
• Francesco Quatraro 
  University of Torino and Collegio Carlo Alberto