Named Entity Recognition for Equipment Fault Diagnosis Based on RoBERTa-wwm-ext and Deep Learning Integration DOI Open Access
Feifei Gao, Zhang Li, Wenfeng Wang

и другие.

Electronics, Год журнала: 2024, Номер 13(19), С. 3935 - 3935

Опубликована: Окт. 5, 2024

Equipment fault diagnosis NER is to extract specific entities from Chinese equipment text, which the premise of constructing an knowledge graph. Named entity recognition for can also provide important data support maintenance support. text has complex semantics, fuzzy boundaries, and limited size. In order this paper presents model based on RoBERTa-wwm-ext Deep Learning network integration. Firstly, uses context-sensitive embeddings sequences. Secondly, context feature information obtained through BiLSTM network. Thirdly, CRF combined output label sequence with a constraint relationship, improve accuracy labeling task, complete task. Finally, experiments predictions are carried out constructed dataset. The results show that effectively identify five types higher evaluation indexes than traditional model. Its precision, recall, F1 value 94.57%, 95.39%, 94.98%, respectively. case study proves accurately recognize input text.

Язык: Английский

Named Entity Recognition for Equipment Fault Diagnosis Based on RoBERTa-wwm-ext and Deep Learning Integration DOI Open Access
Feifei Gao, Zhang Li, Wenfeng Wang

и другие.

Electronics, Год журнала: 2024, Номер 13(19), С. 3935 - 3935

Опубликована: Окт. 5, 2024

Equipment fault diagnosis NER is to extract specific entities from Chinese equipment text, which the premise of constructing an knowledge graph. Named entity recognition for can also provide important data support maintenance support. text has complex semantics, fuzzy boundaries, and limited size. In order this paper presents model based on RoBERTa-wwm-ext Deep Learning network integration. Firstly, uses context-sensitive embeddings sequences. Secondly, context feature information obtained through BiLSTM network. Thirdly, CRF combined output label sequence with a constraint relationship, improve accuracy labeling task, complete task. Finally, experiments predictions are carried out constructed dataset. The results show that effectively identify five types higher evaluation indexes than traditional model. Its precision, recall, F1 value 94.57%, 95.39%, 94.98%, respectively. case study proves accurately recognize input text.

Язык: Английский

Процитировано

2