基于深度学习的地理知识图谱构建方法研究
Research on the Construction Method of Geographic Knowledge Graph Based on BiLSTM-CRF and Bert-BiGRU-Attention Network
任延辉 1苗立志 2黄毅 2汤晟 1张朋东2
作者信息
- 1. 南京邮电大学地理与生物信息学院 南京 210023
- 2. 南京邮电大学地理与生物信息学院 南京 210023;江苏省智慧健康大数据分析与位置服务工程研究中心 南京 210023
- 折叠
摘要
地理知识图谱是提供地理知识服务的关键技术,实现其自动化构建对发展地理人工智能应用具有非常重要的意义.为解决地理知识图谱自动化构建的问题,提出了一种基于BiLSTM-CRF网络提取地理实体和Bert-BiGRU-Attention网络提取方位关系构建领域地理知识图谱的方法.实验结果表明,基于该方法所自动构建的领域地理知识图谱地理实体及其方位信息较为完整,召回率和精确率较高,能够满足知识图谱构建需求,可充分表达现实世界中地理实体及其复杂的方位关系.
Abstract
Geographic knowledge graph is a key technology to provide geographic knowledge services,and its automatic con-struction is of great significance for developing geographic artificial intelligence applications.In order to address the problem of auto-matic construction of geographic knowledge graph,a method is proposed to construct domain geographic knowledge graph based on BiLSTM-CRF network for extracting geographic entities and Bert-BiGRU-Attention network for extracting orientation relations.The experimental results show that the entities and their orientation information in geographic knowledge graph constructed based on the proposed method are relatively complete with high recall and precision.The method proposed in this paper can meet the require-ments of graph construction and can fully describe the geographic entities and their complex orientation relationships in the real world.
关键词
地理知识图谱/BiLSTM-CRF/地理实体/Bert-BiGRU-Attention/方位关系Key words
geographic knowledge graph/BiLSTM-CRF/geographic entities/Bert-BiGRU-Attention/orientation relation-ships引用本文复制引用
基金项目
江苏省"双创博士"项目(CZ032SC20025)
江苏省智慧健康大数据分析与位置服务工程实验室开放基金项目(SHEL221002)
出版年
2024
国家科技期刊平台
NSTL
万方数据