EKGTF: A knowledge-enhanced model for optimizing social network-based meteorological briefings
作者：
Highlights：
• We construct a meteorological briefing formalization module, which consists of three models: the text form judgment model, the formalization words detection model, and the event knowledge guided text formalization (EKGTF) model.
• The EKGTF model with the event knowledge guidance module structure. Such a structure enhances the formalized texts to focus on describing specific meteorological events in the source text. Compared to other baseline models, the EKGTF model achieves the best results.
• The BERT model with meteorological knowledge is fine-tuned to introduce prior knowledge to the EKGTF model. This knowledgeable fine-tuned language model is more sensitive to meteorological events.
• Based on the meteorological briefing formalization module, the service framework for the meteorological briefing formalization is constructed. This framework has been applied to the meteorological briefing overview platform in the CMA Public Meteorological Service Center as an online service.
摘要
•We construct a meteorological briefing formalization module, which consists of three models: the text form judgment model, the formalization words detection model, and the event knowledge guided text formalization (EKGTF) model.•The EKGTF model with the event knowledge guidance module structure. Such a structure enhances the formalized texts to focus on describing specific meteorological events in the source text. Compared to other baseline models, the EKGTF model achieves the best results.•The BERT model with meteorological knowledge is fine-tuned to introduce prior knowledge to the EKGTF model. This knowledgeable fine-tuned language model is more sensitive to meteorological events.•Based on the meteorological briefing formalization module, the service framework for the meteorological briefing formalization is constructed. This framework has been applied to the meteorological briefing overview platform in the CMA Public Meteorological Service Center as an online service.
论文关键词：Event knowledge guided text formalization model,Fine-tuned BERT model,Meteorological event knowledge,Meteorological briefing formalization service framework,Meteorological decision support platform
论文评审过程：Received 30 October 2020, Revised 4 February 2021, Accepted 28 February 2021, Available online 1 April 2021, Version of Record 1 April 2021.
论文官网地址：https://doi.org/10.1016/j.ipm.2021.102564
