电力信息与通信技术2025，Vol.23Issue(9)：42-48,7.DOI:10.16543/j.2095-641x.electric.power.ict.2025.09.06
基于双向融合注意力网络的数据库告警多模态关系抽取
Multi-modal Relation Extraction for Database Alarm Based on Bi-direction Fusion Attention Network
摘要
Abstract
Relation extraction has long been a critical issue in the field of information processing.However,with the development of attention mechanisms and neural networks,most of the current models are designed for unimodal relation extraction,and cannot effectively and reasonably assign weights to features of different modalities in information aggregation.To address these limitations,this paper proposes a multimodal relation extraction model for database alerts based on a bidirectional fusion attention network.This approach highlights explicit and low-level multimodal database alert context by utilizing the bidirectional fusion attention network.To handle the interaction effects between multimodalities,a bottom-up dynamic attention network is proposed to extract implicit multimodal database alert context information.Additionally,a top-down static self-attention mechanism is used to capture and display multimodal interaction information based on the original input at one time.Experimental results on a business system based on Oracle database demonstrate that our model outperforms the current best model,achieving Pre,Rec and F1 scores of 81.5%,83.7%,and 83.4%,respectively.关键词
数据库告警/多模态关系抽取/双向融合注意力网络/多模态交互Key words
multimodal relational extraction/database alarm/attention mechanism/bi-direction fusion attention network/multimodal interactions分类
信息技术与安全科学引用本文复制引用
李济伟,高德荃,冯宝,王至凡,张林锋,卞宇翔..基于双向融合注意力网络的数据库告警多模态关系抽取[J].电力信息与通信技术,2025,23(9):42-48,7.基金项目
国家电网有限公司信息通信分公司自行管理科技项目资助"数据库运行异常事件智能排查与运维知识检索技术研究"(529939220001). （529939220001）
