Knowledge graph-driven question answering for prefabricated building quality management through natural language processing and transfer learning

Purpose Quality compliance checking is an essential process for improving the quality of prefabricated buildings. However, current methods for extracting specific legal compliance often rely on manual operation procedures, which are time-consuming, labor-intensive and prone to errors. In response to the limitation of insufficient knowledge service in this field, this paper proposes a knowledge graph (KG)-driven question answering (QA) method for prefabricated building quality management (KQAP method). Design/methodology/approach This method introduces a novel neural network model that integrates bidirectional encoder representations from transformers (BERT) with an attention-based bidirectional long short-term memory network (AB-BiLSTM). First, a fine-tuned BERT model is employed to perform shallow parsing of regulatory texts using predefined semantic labels. Then, AB-BiLSTM automatically assigns appropriate templates to domain-specific questions, extracting the relevant semantic elements. To avoid terminological inconsistencies between the semantic elements and the corresponding node information in KG, a fuzzy matching method based on text similarity is further employed. Finally, this method enables the automatic generation of query language for KG. Findings Results demonstrate that the method exhibits outstanding performance, with the final hybrid model attaining an accuracy of 95.2%. Furthermore, a prototype QA system is developed, incorporating two key functionalities: quality diagnosis and inspection inquiries. Originality/value This paper presents an effective method for rapid quality diagnosis and timely QA responses, contributing to the improvement of the construction quality management.