面向地理场景的“文—图—景”转换方法研究

发布时间：2018-04-20 17:27

本文选题：地理场景 + 自然语言　；参考：《南京师范大学》2013年博士论文

【摘要】：自然语言和虚拟地理场景是两种重要的地理场景表达手段,在数据的丰富性、表达的直观性、认知的习惯性、应用的普及性等方面各有所长。如何将具有丰富语义的自然语言信息转换到虚拟地理场景中,已成为地理信息学科的热点研究问题。本文依托国家自然科学基金项目“面向自然语言的虚拟地理场景重构方法”,探索自然语言中地理场景信息“结构化、图形化、空间化”过程中的表达机制,实现地理场景从文本到图形和虚拟场景的语义转换。本文的主要研究内容如下： (1)地理场景自然语言描述的结构化表达通过建立地理场景的自然语言描述实例库,分析地理要素和语义的自然语言描述特点,应用框架知识表示法,从“地理要素、空间关系、地理场景结构”三个层次进行地理场景的结构化表达,解决地理场景的自然语言描述与虚拟场景表达的之间数据异构性问题。框架知识表示法不仅能够表达地理场景结构特征,而且保留了地理场景各个地理要素间的语义关联信息。 (2)知识库驱动的地理场景“文—图”转换空间布局图不仅能够表达出地理要素分布的结构特征,而且反映地理场景各地理要素之间的定性语义关系,是一种比自然语言更加直观的地理场景表达方式。通过构建地理场景知识库,包括地理场景概念分类体系、地理常识库、地理场景符号库和地理场景结构模板库等,解决地理场景自然语言描述的语义一致性问题；顾及空间关系语义约束,实现了地理要素的空间“粗布局”；提出了基于蚁群算法的地理场景空间布局优化方法,形成人机辅助的地理场景“文—图”转换机制。 (3)基于相似性匹配的地理场景的“文—景”转换通过地理概念相似度、空间关系相似度和空间结构相似度计算,实现了基于多层次空间相似性匹配的自然语言与虚拟地理场景转换方法。实验结果表明,地理场景的自然语言描述与虚拟地理场景在表达方式上存在较大差异,两者之间的转换实际上是一个寻找空间和语义相似解的过程。但是,自然语言地理场景结构特征的描述信息较为匮乏,借助地理场景的空间布局图,能够更加有效地实现地理场景的“文—景”转换。
[Abstract]:Natural language and virtual geographical scene are two important means of geographical scene expression, which have their own advantages in the richness of data, the intuitiveness of expression, the habit of cognition, the popularization of application, etc. How to transform natural language information with rich semantics into virtual geographic scene has become a hot topic in the field of geographic information. Based on the National Natural Science Foundation project "Virtual Geographic scene Reconstruction method for Natural language", this paper explores the expression mechanism in the process of "structured, graphical and spatial" of geographic scene information in natural language. The semantic transformation of geographical scene from text to graph and virtual scene is realized. The main contents of this paper are as follows: 1) structured representation of natural language description of geographical scenes By establishing the natural language description case base of geographical scene, analyzing the characteristics of geographical element and semantic natural language description, applying the frame knowledge representation, this paper analyzes the relationship between geographical elements and space. In order to solve the problem of data heterogeneity between the natural language description of geographical scene and the representation of virtual scene, the three levels of geographical scene structure are structured expression of geographical scene. Frame knowledge representation can not only express the structural features of geographical scenes, but also retain the semantic association information among geographical elements. Knowledge-driven "text-graph" transformation of geographical scenes Spatial layout map can not only express the structural characteristics of geographical elements distribution, but also reflect the qualitative and semantic relationship between geographical scenes. It is a more intuitive way to express geographical scenes than natural language. The semantic consistency of geographical scene natural language description is solved by constructing geographical scene knowledge base, including geographical scene concept classification system, geographical common sense database, geographical scene symbol database and geographical scene structure template database. Taking into account the semantic constraints of spatial relations, the spatial "rough layout" of geographical elements is realized, and the optimization method of geographical scene spatial layout based on ant colony algorithm is proposed, which forms a human-computer aided "text-graph" transformation mechanism. "text-scene" transformation based on similarity matching By calculating the similarity of geographical concept, spatial relationship and spatial structure, the transformation method of natural language and virtual geographic scene based on multi-level spatial similarity matching is realized. The experimental results show that there is a great difference between the natural language description of geographical scenes and the virtual geographical scene in terms of expression, and the transformation between them is actually a process of searching for spatial and semantic similarity solutions. However, the description of natural language geographical scene structure features is relatively scarce. With the help of spatial layout map of geographical scene, the transformation of geographical scene "text-scene" can be realized more effectively.
【学位授予单位】：南京师范大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：P208

【参考文献】