IEEE Access (Jan 2024)

Research on Joint Extraction Method of Elevator Safety Risk Control Knowledge Based on Multi-Perspective Learning

  • Suli Hao,
  • Fenfen Shi

DOI
https://doi.org/10.1109/ACCESS.2024.3481266
Journal volume & issue
Vol. 12
pp. 159488 – 159502

Abstract

Read online

Knowledge graph visualizes the knowledge domain by describing various entities and concepts existing in the real world and the relationship between them, which is the key to intelligent recommendation and intelligent decision-making. However, the low accuracy of knowledge extraction has become a major bottleneck in the construction of high-quality knowledge graphs. The reason for this is because of the problem of highly overlapping triples and long-distance dependence between triples in each domain knowledge. Taking elevator safety risk control as the research object, this paper proposes a joint extraction algorithm with multi-perspective learning. Based on CasRel model, for the limitations of Casrel model, in the feature extraction stage, multi-head attention mechanism, bidirectional gated loop unit and convolutional neural network are introduced to extract the features output from BERT layer respectively. The purpose of extracting data features from multiple perspectives such as bidirectional perspective, local perspective and global perspective is realized. In this study, 5765 labeled elevator safety risk text data sets were divided into training sets and test sets according to the ratio of 4:1. The data set is used to verify the CasRel model before and after the improvement. The experimental results show that the values of P (accuracy rate), R (recall rate) and F1 of the improved CASREL model are 82.06%, 79.53% and 80.78%, respectively, which are 6.60%, 9.32% and 8.03% higher than those of the traditional CASREL model. This proves the validity of the improved CasRel model (BERT-BiGRU-MHA-CNN-CasRel). At the same time, it shows that the improved CasRel model can better complete the triplet extraction task, and can better complete the triplet extraction work. In addition, it is also confirmed that this method can extract elevator safety risk prevention and control knowledge more accurately, which provides data support for intelligent decision-making of elevator safety risk prevention and control based on knowledge graph, and also provides method support for knowledge extraction in other fields.

Keywords