Leveraging Language Embeddings for Missing Data Imputation in Multilingual Digital Product Passports

Digital Product Passports (DPPs) have emerged as an essential solution to enhance product transparency, traceability, and sustainability across various supply chains. DPP serves as the key element of Circular Economy for the entire product life cycle: from production and consumption to waste management and the secondary materials market. In their infancy, DPPs suffer from various issues, mainly incomplete information, which hampers their utility and reliability. This becomes more challenging in multilingual environments where linguistic diversity further complicates data consistency and completeness. This paper proposes a framework that uses multilingual language embeddings to address the problem of missing data imputation in DPPs. By employing state-of-the-art natural language processing models, such as mBERT, we develop a framework that generates language-agnostic embeddings to predict and fill missing data fields across multiple languages within DPPs. Specifically, the proposed framework addresses data incompleteness in DPPs by learning the contextual relationships and semantic similarities between present and absent data entries. Our experimental results demonstrate significant efficiency in data imputation accuracy, as validated across diverse languages.