The growth in the number of industries aiming at more sustainable business processes is driving the use of the European Waste Catalogue (EWC). For example, the identification of industrial symbiosis opportunities, in which a user-generated item description has to be annotated with exactly one EWC tag from an a priori defined tag ontology. This study aims to help researchers understand the perils of the EWC when building a recommender system based on natural language processing techniques. We experiment with semantic enhancement (an EWC thesaurus) and the linguistic contexts of words (learned by Word2vec) for detecting term vector similarity in addition to direct term matching algorithms, which often fail to detect an identical term in the short text generated by users. Our in-depth analysis provides an insight into why the different recommenders were unable to generate a correct annotation and motivates a discussion on the current design of the EWC system.
- European waste catalogue (EWC)
- Industrial symbiosis
- Recommender system
- Tag recommendation
- Circular economy