【作者】 李晗静；

【导师】 李生；

【作者基本信息】 哈尔滨工业大学 ， 计算机应用技术， 2007， 博士

【摘要】 空间概念建模是文景转换研究中一个十分基本的问题。研究自动、有效的自然语言描述空间概念建模方法,对于自然语言理解及文景转换研究具有重要的理论意义和实用价值。本文以自然语言描述空间关系为研究对象,提出了一整套基于自然语言描述的空间概念建模方案。作为概念建模的一个重要环节,研究了空间本体库构建以及空间关系抽取技术。在此基础上,研究并且实现了基于两步法的物体空间摆放。基于t检验的目测评价方法对空间概念建模原型系统的场景进行了评价,验证了小规模空间本体库的有效性,以及本文提出的基于自然语言描述空间概念建模方案的可行性和重要价值。具体地讲,本文从如下几个方面进行了研究:1.研究了空间本体库构建技术。分析了空间本体库构建需要解决的关键问题,提出了基于SUMO的空间本体概念、关系定义以及基于半自动化实例获取方法,有效的解决了本体实例化以及间接相关问题。在此基础上,综合利用汉语语义库、图形资源构建了小规模的空间本体库。2.研究了篇章级空间关系抽取技术。分析了目前信息抽取方法特点以及待解决问题,提出了基于线性分类器的空间关系抽取方法。提出汉语描述空间关系的形式化描述,以及抽象空间关系抽取为二值分类问题。在词性标注的基础上,结合词性、空间语义等特征,分层次地实现了篇章级空间关系抽取。该方法充分利用了汉语词性标注研究成果,避开了汉语句法分析以及语义分析的难题,有效地实现了空间认知概念的抽取,同时保证了正确率。3.在小规模空间本体库基础上研究了物体空间摆放方法。分析了目前概念建模方法特点以及待解决关键问题,提出基于区域确定实现物体空间摆放方法。区域确定是基于小规模空间本体库和聚类策略确定实体摆放区域方法,平滑地实现定性信息到定量信息的转换,离散自然语言系统到连续图形系统的过渡,同时有效地缩小了物体摆放搜索空间。以此为前提,提出基于最优保持简单遗传算法的物体精确摆放方法,该方法保证了搜索全局最优性。该方法有效地降低了问题难度,同时保证了物体空间摆放的合理性,为概念建模提出了一个崭新的思路。4.建立了基于自然语言描述的空间概念建模原型系统。探讨了空间概念建模系统的基本框架、理解可视化流程,以及提出了基于t检验目测评价方案。基于t检验目测评价方法,通过对系统可视化场景的评价验证了基于自然语言描述的空间概念建模方案的有效性和可行性。更多还原

【Abstract】 Spatial conceptual model is a fundamental problem in the text-to-scene conversion tasks. The research on effective automatic spatial conceptual model which is described in natural language is of great theoretical and practical significance. This paper presents a method for automatic spatial conceptual model based on natural language descriptions. As a crucial step, it first studies methods for the definition and creation of the spatial ontology and the extraction of spatial information. Then, based on these, the layout of 3D objects based on two steps is researched and realized. The prototype of sptial conceptual model is evaluated based on t-test we proposed. The evaluation suggested methods are usedful in real model.This thesis is arranged as follows:1. The definition and acquisition of the spatial ontology is studied. On analysis the issues in spatial ontology creation, the methods of the definition of spatial concepts and relations based on SUMO and the acquisiton of spatial ontology based on clustering are proposed. The methods solves the difficulty of the spatial ontology instantiation automaticly and the problem of indirect association in cluster process. On the basis of these methods, the small spatial ontology is created on the resources including HowNet and graphics.2. The extraction of spatial relationships in text level is studied. On analysis the issues in spatial information extraction, we propose the method extracting spatial relationships based on binary-classification. At first, we give definitions on spatial relationships in Chinese formally and moded the extraction of the spatial relationship as a binary classification problem. And then, on the basis of part-of-speech tagging, the extraction of spatial relationships in text level is realized step by step on the features on part-of-speech and spatial semantics. The method takes advantage of the results on part-of-speech tagging in Chinese, keep away from the usage of syntactic parser and sematic parser, and realized the extraction on the congnitive level with the good performance.3. Based on the small spatial ontology, the layout of 3D objects is studied. We proposed the layout of 3D objects on the qualification of the layout area on the analysis the issues on objects layout. The qualification of the layout are is based on spatial ontology and clustering, crucial to turn the qualitative information into quantitative information, and the bridge between the symbolic system and the numeric system and reduces the search space for the exact layout. After that, exactly layouting of 3D objects based on OMSGA is done. The OMSGA can ensure to search the global optimization. The method reduces effectively the difficulty of objects layout and resolve objects layout deeply and soundly. It opens new way to conceptual model.4. A prototype of Chinese spatial conceptual model system is implemented. The frame of the system and the modeling procedure are discussed. The evaluation based on t-test is proposed. Evaluation on the visualization results demonstrates the good performance of the schema of spatial conceptual model.更多还原