基于多任务学习的电力文本信息抽取  

Power text information extraction based on multi-task learning

作　　者：纪鑫 武同心 余婷 董林啸 陈屹婷 米娜 赵加奎 JI Xin;WU Tongxin;YU Ting;DONG Linxiao;CHEN Yiting;MI Na;ZHAO Jiakui(Big Data Center of State Grid Corporation of China,Beijing 100031,China;School of Computer Science and Engineering,Beihang University,Beijing 100191,China)

机构地区：[1]国家电网有限公司大数据中心,北京100031 [2]北京航空航天大学计算机学院,北京100191

出　　处：《北京航空航天大学学报》2024年第8期2461-2469,共9页Journal of Beijing University of Aeronautics and Astronautics

基　　金：基于图神经网络和图深度学习的电力知识抽取技术研究(52999021N005)。

摘　　要：为提升电力系统故障文本在实际业务场景中的分析处理速度,提出基于预训练与多任务学习的电力故障文本信息自动抽取模型。利用预训练模型学习电力文本词语的上下文信息,挖掘词语的一阶和二阶融合特征,增强特征的表示能力,利用多任务学习框架结合命名实体识别和关系抽取2个任务的学习,实现实体识别和关系抽取的互相补充和互相促进,进而提高电力故障文本信息抽取的性能。通过对某电力网数据中心的日常业务数据进行模型验证,与其他模型相比,所提模型提高了电力故障文本实体识别和关系抽取的准确率和召回率。In order to improve the analysis and processing speed of power system fault text in actual business scenarios,a power fault text information extraction model based on pre-training and multi-task learning was proposed.The pre-training model was used to learn the context information of power text words.The first-order and secondorder fusion features of words were mined,which enhanced the representation ability of features.The multi-task learning framework was used to combine named entity recognition and relation extraction,which realized the mutual supplement and mutual promotion of entity recognition and relationship extraction,so as to improve the performance of power fault text information extraction.The model was verified by the daily business data of a power data center.Compared with other models,the proposed model’s accuracy and recall of power fault text entity recognition and relation extraction were improved.

关 键 词：电力故障 预训练 多任务学习 实体识别 关系抽取 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]