Semi-automatic mapping with SBERT for ontology-based interoperability in construction data systems

The construction industry faces significant challenges in achieving interoperability among software systems due to disparate data models, inconsistent semantics, and varying standards. These issues lead to increased project costs, operational inefficiencies, and delays while also creating data silos that hinder collaboration and impair decision-making. This paper introduces an adaptive ontology-based framework employing SBERT-based automatic mapping to address these challenges. SBERT (Sentence-BERT) is a neural network model optimised for sentence embeddings, enabling precise semantic similarity comparisons between textual data. The approach integrates domain-specific knowledge and contextual fine-tuning to enhance the accuracy and relevance of semantic mapping. The framework uses an adaptive ontology as a standard vocabulary to bridge semantic gaps. It harmonises Building Information Modeling and Environmental Product Declaration data from platforms like iTwins, Building Transparency, and Moata Carbon Portal. A five-stage process—data acquisition, ontology generation, semantic mapping, ontology creation, and validation—guides development. The ontology evolves through user feedback, automation, and new domain knowledge, ensuring scalability. This study uses three diverse datasets (JSON, XML, CSV) to generate a generic carbon report, enabling seamless identification of missing values, definitions, and data structures. Validated with data from the TransPennine Route Upgrade project, the framework tackles real-world interoperability challenges in construction and carbon management. Future work aims to refine heatmap interpretability and expand semantic metrics.