基于BERT的遥感应用领域知识图谱自动构建技术  

Automatic Construction Technology of Remote Sensing Application Domain Knowledge Graph Based on BERT

作　　者：李峰[1] 王琼洁 韦二龙 刘义贤 陈旭[1,3] LI Feng;WANG Qiong-jie;WEI Er-long;LIU Yi-xian;CHEN Xu(CETC key laboratory of aerospace information applications,Shijiazhuang 050081,China;China Center for Information Industry Development,Beijing 100846,China;School of Computer Science,Wuhan University,Wuhan 430072,China)

机构地区：[1]中国电子科技集团公司航天信息应用技术重点实验室,河北石家庄050081 [2]中国电子信息产业发展研究院,北京100846 [3]武汉大学,湖北武汉430072

出　　处：《中国电子科学研究院学报》2021年第7期645-653,共9页Journal of China Academy of Electronics and Information Technology

基　　金：中国电子科技集团公司航天信息应用技术重点实验室开放基金项目(SXX18629X015)。

摘　　要：随着对地观测技术的快速发展,卫星遥感影像应用场景也日益丰富。然而,由于遥感数据应用门槛相对过高,专业知识储备要求相对严格,遥感应用领域知识缺乏系统化、结构化表示,严重制约了研究者和从业人员在特定行业中快速开发和应用遥感数据资源。文中将高质量的遥感应用领域学术论文和学术专著作为高质量知识库来源,使用引入损失函数Lenloss的BERT-BiLSTMCRF模型作为命名实体识别模型,抽取出卫星实体、传感器实体、分辨率实体、任务实体以及方法实体,进一步将同一篇学术论文摘要中不同实体之间的共现行为用于定义实体之间的关系,抽取出不同实体之间的关系。最后,利用图数据库Neo4j实现遥感应用领域知识图谱的存储管理。With the rapid development of earth observation technology,satellite remote sensing image application scenarios are increasingly rich. However,due to the threshold in the field of remote sensing application is very high,this domain knowledge is still short of a systematic and structural representation,which limits its application for researchers and practitioners. This paper selects academic thesis and monograph as the source of domain knowledge,uses BERT-BiLSTM-CRF with Lenloss as named entity recognition model,which extracts satellite entity,sensor entity,resolution entity,task entity and method entity. The co-occurrence of different entities in the same paper is defined as relationship,which is relation extraction between named entities. Finally,Neo4 j is used to realize the storage and management of remote sensing application domain knowledge graph.

关 键 词：预训练 领域知识图谱 遥感应用 命名实体识别 

分 类 号：F49[经济管理—产业经济]