Joint Extraction of Equipment Fault Text Entity Relations Based on BERT-GPLinker
摘要
The continuous application of high and new technologies has rapidly increased the complexity of equipment and the difficulty of maintenance support. Addressing issues such as the low utilization of unstructured data and complex entity relationships in equipment fault diagnosis texts, this paper presents an entity-relation joint extraction model for equipment fault diagnosis texts. By integrating a BERT encoder, attention mechanism, global pointer network, and token-pair linking joint extraction network, we designed an improved BERT-GPLinker. This enhancement achieves precision, recall, and F1-scores of 93.83%, 93.08%, and 93.45% respectively. Experimental results validate the model’s effectiveness and accuracy.