Acceso abierto

Real-Time Extraction of News Events Based on BERT Model

Yuxin Jiao

Yuxin Jiao

School of Computer Science and Engineering, Xi'an Technological UniversityXi’an, China
Buscar este autor en
Sciendo|Google Scholar
Jiao, Yuxin
 y 
Li Zhao

Li Zhao

School of Computer Science and Engineering, Xi'an Technological UniversityXi’an, China
Buscar este autor en
Sciendo|Google Scholar
Zhao, Li
  
30 sept 2024
Real-Time Extraction of News Events Based on BERT Model's Cover Image
Acerca de este artículo
Artículo anterior
Artículo siguiente
Cite
Descargar portada
Publicado en línea: 30 sept 2024
Páginas: 24 - 31
DOI: https://doi.org/10.2478/ijanmc-2024-0023
Palabras clave
© 2024 Yuxin Jiao et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

Event Extraction Process Map
Event Extraction Process Map

Figure 2.

Pre-training and Fine-Tune process
Pre-training and Fine-Tune process

Figure 3.

Graph structure of CRFs for linear chain conditional random fields
Graph structure of CRFs for linear chain conditional random fields

Figure 4.

Extracted event output structure, including event types and argument roles
Extracted event output structure, including event types and argument roles

Figure 5.

Comparison of P-value, R-value and F1-value of LSTM, BiLSTM and BERT-CRF models. P for Precision, R for Recall
Comparison of P-value, R-value and F1-value of LSTM, BiLSTM and BERT-CRF models. P for Precision, R for Recall

Figure 6.

Comparison of P-value, R-value and F1-value of BERT, RoBERTa and ALBERT models Comparison of P, R and F1 values. P for Precision, R for Recall
Comparison of P-value, R-value and F1-value of BERT, RoBERTa and ALBERT models Comparison of P, R and F1 values. P for Precision, R for Recall

Experimental Results I

Module P R F1
LSTM 34.2% 40.6% 37.1%
BiLSTM 58.1% 56.2% 57.1%
BERT-CRF 76.5% 76.9% 76.7%

Experimental Results II

Module P R F1
BERT 76.50% 76.90% 76.70%
RoBERTa 77.60% 77.10% 77.30%
ALBERT 78.10% 77.30% 77.70%
Idioma:
Inglés
Calendario de la edición:
4 veces al año
Temas de la revista:
Informática, Informática, otros
RSS Feed de revista