Using Natural Language Processing to monitor circular activities and employment

In Europe, NACE codes are used for the official classification of sectors, however, the circular economy is not sufficiently captured in this classification. Therefore, this paper improves previous attempts for defining circular activities and jobs by web scraping techniques applied to each company in Belgium. We analyze their first, second, and third official NACE codes and compare these to the NACE codes they should have been allocated to according to the web scraping data. Subsequently, we calculate circularity scores for every sector to construct an indicator for the number of circular companies and jobs. The results show that the number of circular companies is lower than the baseline from official statistics when we only consider the companies' first and main NACE code. The estimates are higher than the baseline when we also take the second and third NACE codes into account and the estimated number of circular jobs is far higher than the baseline. This research upgrades previous classifications of circular sectors and demonstrates how web scraping and novel data might improve our understanding and capacity to build data. Based on the results in this paper, we recommend a uniform data collection such as reporting standards, and an inclusion of all circular strategies in sectoral classifications.