重庆科技大学学报（自然科学版）2025，Vol.27Issue(3)：78-85,8.DOI:10.19406/j.issn.2097-4531.2025.03.008
基于提示学习的突发事件新闻文本层次多标签分类方法研究
Research on Hierarchical Multi-Label Classification Method for Emergency News Texts Based on Prompt Learning
汪波 1余茂楠 1唐伟 1张万宏 1马代强 1邓松2
作者信息
- 1. 重庆天然气储运有限公司,重庆 401139
- 2. 西南油气田分公司重庆气矿,重庆 401120
- 折叠
摘要
Abstract
The classification of emergency events serves as a crucial prerequisite for emergency response operations,directly determining the speed and effectiveness of response measures.To effectively address the problem of classifi-cation accuracy caused by imbalanced categories in emergency news texts,this study proposes a hierarchical multi-label classification method for emergency news texts based on prompt learning.By constructing prompt templates upon the ERNIE pre-trained model,we utilize the masked language model to train predicted labels and achieve la-bel mapping to match existing classification labels.This approach effectively mitigates challenges stemming from limited annotated data and data imbalance in the emergency domain.Experimental results demonstrate that the pro-posed model achieves accuracy and macro-F1 scores of 0.973 9 and 0.933 7 respectively,outperforming baseline models such as ChineseBERT and PET.关键词
提示学习/突发事件/新闻文本分类/层次多标签Key words
prompt learning/emergency events/news texts classification/hierarchical multi-label分类
信息技术与安全科学引用本文复制引用
汪波,余茂楠,唐伟,张万宏,马代强,邓松..基于提示学习的突发事件新闻文本层次多标签分类方法研究[J].重庆科技大学学报（自然科学版）,2025,27(3):78-85,8.
