NLP-based Data-Enrichment for Building Management

One of the main challenges in building management is dealing with the large amount of unstructured data produced throughout the asset's life cycle. At handover, Building Information Models often provide low-quality, incomplete data, necessitating extensive manual rework to elicit information from many sources. To minimize this manual rework, this study proposes an automated Information Extraction (IE) procedure, applied to the design and construction documents to extract information, enrich a model (COBie format), and maximize the transfer of structured data to the client. The proposed approach is based on Natural Language Processing (NLP) and adopts Transformer-based Named Entity Recognition (NER) and Relation Extraction (RE) methods for IE. It was evaluated and achieved good performance, with average F1 scores of 0.73 for NER and 0.91 for RE, representing a step toward a reliable tool for an enhanced data handover process. © 2024 European Council on Computing in Construction.