面向虚拟地理环境的自然语言问答式知识搜索研究

The knowledge search based on natural language question answering is a hot and difficult point in the cross study of the two fields of virtual geographic environment and artificial intelligence. At present, knowledge search for virtual geographic environment is mostly based on keyword and rule template, which is difficult to meet the needs of deep interaction and intelligent service for intelligent virtual geographic environment.. Therefore, this project supported by computational linguistic theory and methods, using knowledge graph technology, researches further on virtual geographic environment knowledge representation and intelligent service. At first，construct the multi-hierarchy semantic model of virtual geographic environment, according to the characteristics of multi-scale virtual geographic environment, on the basis of linguistic theory. Secondly, explore approaches for geographic knowledge extraction to construct geographic knowledge base. In addition, study the space syntax analysis model and improve the algorithms of natural language text geographical entity recognition and links, considering the spatial relationship of the sentence characteristics and the constraints between spatial relationship patterns and geographical elements. Finally, the experimental scheme is taken for geographic knowledge extraction and interactive knowledge search.. The research of this project will help to improve the interaction and intelligent service level of virtual geographic environment system, and promote the construction of virtual geographic environment knowledge engineering.

自然语言问答式知识搜索是虚拟地理环境与人工智能两大领域交叉研究的热点与难点。目前虚拟地理环境知识搜索多是基于关键词、规则模板的方法，难以满足智能虚拟地理环境深度交互和智能服务的需要。. 对此，本项目以计算语言学相关理论与方法为支撑，利用知识图谱相关技术，围绕虚拟地理环境事实型知识表示和智能服务展开研究：1)针对虚拟地理环境多尺度的特点，借鉴语言学相关理论，构建虚拟地理环境多层次语义解析模型；2)研究面向虚拟地理环境场景的地理知识抽取方法，构建地理知识库；3)应用计算语言学空间关系句式特征和地理要素与空间类型的依存关系，研究地理特征约束的空间句法分析模型，改进自然语言短文本地理实体识别与链接算法；4）设计实验方案对地理知识抽取和问答式知识搜索进行实验。. 本课题的研究将有助于提高虚拟地理环境系统的交互性和智能化服务水平，推动虚拟地理环境知识工程的建设进程。

智能化虚拟地理环境是VGE发展的必然趋势，地理知识图谱是地理认知智能的必经之路，利用地理知识图谱实现地理场景智能交互和场景推理在大数据条件下具有显著优势。本项目以虚拟地理环境智能服务为出发点，提出了一种知识增强的虚拟地理环境服务框架，构建了地理环境空间到语义的多层次解析模型，研究了面向虚拟地理环境的多模态知识图谱形式化表达方法，探索了遥感影像目标知识分类及知识表示，提出了大规模遥感影像目标知识图谱构建方法，研发了基于地理知识图谱的虚拟场景智能交互系统，并在问答式场景交互、新冠病患时空分析上进行了实验验证。通过本项目的研究，形成了一种面向虚拟地理环境的知识表示和地理知识图谱构建方法，有助于推动虚拟地理环境的智能化发展。本项目执行期间，发表学术论文6篇，申请发明专利2项，软件著作权登记2项，培养研究生5名（其中博士生2名，硕士生3名）。研究成果成功应用于国防军事、医疗卫生等领域，未来在地理空间智能服务和军事智能认知领域具有很强的应用潜力。