BEKG: A built environment knowledge graph
Xiaojun Yang,Haoyu Zhong,Zhengdong Wang,Penglin Du,Keyi Zhou,Heping Zhou,Xingjin Lai,Yik Lun Lau,Yangqiu Song,Liyaning Tang
DOI: https://doi.org/10.1080/09613218.2023.2238851
2023-08-02
Abstract:In recent years, the digitalization of the built environment has progressed rapidly due to the development of modern design and construction technologies. However, the need for extensive professional knowledge in this field has not been met by practitioners and scholars. To address this problem, a study was conducted to build a knowledge graph in the built environment domain, which stores entities and their connections in a graph data model. To achieve it, this research collected more than 80,000 paper abstracts from the built environment domain. To ensure the accuracy of entities and relationships in the knowledge graph, two well-annotated datasets were created with 29 types of relationships, each containing 2000 and 1450 instances, respectively, for Named Entity Recognition (NER) and relationship extraction (RE) tasks. Two BERT-based models were trained on these datasets and achieved over 85% accuracy in both tasks. Using these models, over 200,000 high-quality relationships and entities were extracted from abstract data. This comprehensive knowledge graph will help practitioners and scholars better understand the built environment domain.
construction & building technology