Joint triple Extraction for construction regulatory documents using graph neural networks

The construction industry depends on regulatory documents, to guarantee the consistency and applicability of their operations, which is crucial for maintaining standards and protocols. Information extraction is a time-consuming and labor-intensive process, which aims to extract the relevant data from documents for definite operation. Extracting information and assigning it to a particular rule requires understanding the semantics of the sentences, essential for ensuring code compliance. One approach to address this concern is to parse the unstructured text into triples by extracting entities and their relation. This process enables the identification of subjects, objects, and relationships that provide a comprehensive understanding of the sentence. This paper proposes a joint entity and relation extraction model based on a graph neural network. The model is trained on construction regulatory documents sourced from Eurocode for design. The dataset undergoes preprocessing and labeling in the form of triples to assess the model’s performance. The results demonstrate that the model can effectively predict the sentence triples. The model’s prediction of sentence triples indicates the ability to capture complex semantic relationships within textual data, which can be transformed into a comprehensive knowledge representation.