Bridging trends and science: Cluster analysis for topic extraction within the circular economy

The transition towards a more sustainable future has become pivotal for the 21st century. In this sense the shift to a circular economy presents one option to meet the ""Sustainable Development Goals"". The circular economy seeks to minimise waste generation and material inputs through eco-design, recycling and reusing of products. Here, numerous trends are emerging. Identifying and assessing trends and their scientific basis is important for companies, researchers and policy-makers to support the transition towards a circular economy. Fundamental to this task is the detection of weak signals and other indicators. Bridging information from trend data and scientific publications is a promising opportunity to identify and analyse the scientific basis of trends and to assess their importance in supporting the decision-making process. We collect title and abstract of publications dealing with circular economy and use the TrendOne database to collect similar data on trends associated with this topic. Two approaches are applied and compared. First, we bridge the two data sources using the doc2vec approach and apply an appropriate clustering approach to identify topics within this data set. The second approach is to cluster the data sets separately by doc2vec and e.g. k-means. We bridge the clusters by measuring semantic similarities with doc2vev. For both approaches we analyse the development of the clusters to identify the clusters evolutionary path. Current results reveal that trend data and scientific publications complement each other and clustering approaches based on semantic similarities provide insights into the scientific basis of trends and their development.