摘要
优质的设备故障领域本体需要实例之间拥有准确的关系。传统的基于神经网络的实例关系抽取方法容易使意义相近的关系误分,且由于故障领域文本中实例间隔较远,使用该类方法抽取故障领域实例关系效果不佳。针对上述问题,基于分层结构提出了一种从非结构化文本中抽取设备故障领域实例关系的方法。第一层通过对实例对中实例类型的识别,进行初步关系抽取;第二层使用融合自注意力机制的BiLSTM模型对包含实例对的句子进行解析,进行精确分类;通过统计的方法完成实例关系抽取。实验结果表明,此方法与现有的基于神经网络的实例关系抽取方法相比,取得了较好的精确度与召回率,提高了故障本体实例关系抽取的效果(F1值)。
High quality equipment failure domain ontologies need accurate relations between instances.However,the traditional relation extraction methods based on neural network may mix up closely-related relations.Because the distances between instances are far apart in the equipment failure domain texts,there exists difficulty in the task of instances relation extraction.To solve the above problems,this paper proposes a method to extract equipment failure domain ontology instance relations from unstructured text based on a hierarchical structure.In first layer,by identifying the classes of the instances,relation between instances was extracted preliminarily.In second layer,BiLSTM model based on joint self-attention mechanism was used to parse sentences containing instance pairs for accurate classification.The relation extraction was completed by statistical method.The research results show that this method has achieved better precision and recall than the existing instances extraction methods based on neural network,and improved the effectiveness of failure ontology instance relationship extraction(F1 score).
作者
葛天一
杨长春
陈延雪
周婷
Ge Tianyi;Yang Changchun;Chen Yanxue;Zhou Ting(Aliyun School of Big Data,Changzhou University,Changzhou 213016,Jiangsu,China;School of Microelectronics and Control Engineering,Changzhou University,Changzhou 213016,Jiangsu,China)
出处
《计算机应用与软件》
北大核心
2023年第8期72-79,共8页
Computer Applications and Software
基金
江苏省研究生科研与实践创新计划项目(KYCX19_1771)
赛尔网络下一代互联网技术创新项目(NGII20180706)。
关键词
本体构建
设备故障
关系抽取
分层结构
自注意力机制
双向长短期记忆模型
Ontology construction
Equipment failure
Relation extraction
Hierarchical structure
Self-attention mechanism
BiLSTM