Integrating deep learning and multi-attention for joint extraction of entities and relationships in engineering consulting texts

Abstract

While traditional manual knowledge management methods indicate the intelligent approach in the whole-process engineering consulting, related studies like NLP technologies still demonstrated the feasibility and difficulties in processing the complex unstructured long-text consulting knowledge text. To optimize, by firstly incorporating multi attention mechanisms to realize complex long-text knowledge processing and subsequently integrating optimized BERT model RoBRETa and CASREL model for jointly extracting entities and relationships from texts, this paper proposes a LF-CASREL model to optimizes existing knowledge management techniques. Validation experiment with a knowledge graph and question-answering interactions after jointly extraction through LF-CASREL with a precision of 88.89 %, a recall of 77.25 %, and a F1 score of 68.99 % under practical random noise influence demonstrates the practicality of the proposed method. Overall, the proposed LF-CASREL is convenient and beneficial for project managers, engineering consultants, and decision-makers in deeper understanding and management of whole-process engineering consulting, providing valuable insights for future research.