基于MRC的设备故障命名实体识别方法
MRC-BASED METHOD FOR EQUIPMENT FAULT NAMED ENTITY RECOGNITION
徐鹏 1龚伟 2宋俊典3
作者信息
- 1. 公安部第三研究所 上海 200031
- 2. 上海地铁第一运营有限公司 上海 200003
- 3. 上海计算机软件技术开发中心 上海 200112
- 折叠
摘要
命名实体识别是一种有效的设备运行日志分析方法,不仅提高了故障检测的准确度,而且为智能运维策略的优化提供了强有力的支持.鉴于设备运行日志的专业性和复杂性,提出一种基于机器阅读理解的设备故障命名实体识别方法,该方法通过将特定的实体类别转化为 自然语言查询,并将实体类别信息融合到这些查询中,有效地克服了传统方法在标签语义信息上的不足,并在实体边界定位的准确性上取得了显著提升.实验表明该方法在设备故障命名实体识别的准确性和有效性方面明显优于现有的基线方法.
Abstract
Named entity recognition plays a crucial role in analyzing equipment operation logs,not only enhancing the accuracy of fault detection but also providing strong support for the optimization of maintenance strategies.Given the professionalism and diversity of equipment operation logs,this paper proposes a fault entity recognition method based on machine reading comprehension.This method transformed specific entity categories into natural language queries and integrated entity category information into these queries,which effectively overcame the deficiencies of traditional methods in label semantic information and significantly improved the accuracy of entity boundary positioning.Experiments conducted on a dataset for equipment fault entity recognition show that this method excels in accurately identifying equipment fault entities,significantly outperforming existing baseline methods.
关键词
机器阅读理解/命名实体识别/设备故障Key words
Machine reading comprehension/Named entity recognition/Equipment fault引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据