摘要
面向生态环境领域,提出了一种利用文献文本的领域知识图谱自动构建方法。首先利用机器学习技术抽取生态环境领域的关键词和关键关系;再结合统一建模语言系统构建生态环境领域知识图谱模式层,并以该模式层为指导建立生态环境领域语料库;然后采用深度学习方法从文本文献中自动抽取关键信息,通过来源指数和校验次数保证抽取知识的置信度;最终实现知识图谱的自动化构建与可视化展示。结果表明,利用2000万字规模的文献文本可得到含29490条知识三元组的知识图谱,利用Transformer模型抽取的总体精度为96.27%。
In this paper,we proposed a knowledge graph construction method in the ecology and environment field based on literature texts.Firstly,we applied machine learning and modeling language system to structure the model layer of domain knowledge graph.Then,we utilized a deep learning technique to extract key information from literature texts automatically.In order to guarantee the accuracy of extracted knowledge,we used the source index and verification times to ensure the confidence of extracted knowledge.Through experiments,we found that the knowledge graph containing 29490 knowledge triplets was obtained by using 20 million words of literature texts,and the total extraction accuracy by using Transformer model was 96.27%.
作者
王天一
孟小亮
张华
WANG Tianyi;MENG Xiaoliang;ZHANG Hua(School of Remote Sensing Information Engineering,Wuhan University,Wuhan 430079,China)
出处
《地理空间信息》
2023年第1期14-19,共6页
Geospatial Information
基金
国家重点研发计划资助项目(2020YFD1100203)
国家自然科学基金面上基金资助项目(41971352)。
关键词
领域知识图谱
语义标注
深度学习
机器学习
生态环境
domain knowledge graph
semantic annotation
deep learning
machine learning
ecology and environment
