基于概念图的动画剧本生成的研究

发布时间：2018-05-15 17:30

本文选题：概念图 + 概念关系抽取　；参考：《西北大学》2013年硕士论文

【摘要】：动漫产业的迅速发展带动了经济增长并丰富了人们的生活,而传统的动画生成需要大量人工繁琐的劳动,动画的自动生成可有效的将人力解放出来,节省大量资源。动画剧本的生成是动画自动生成中的关键环节,是连接故事文本与动画场面的桥梁,如何抽取动画剧本所需的剧本元素,并选取合适的方式将剧本组织起来是解决问题的关键。对于动画剧本生成的研究,国外已有相当数量的研究成果,而由于中文构词的特殊性,国内的研究还有所欠缺。本文运用了自然语言理解技术,将剧本生成系统分成了三个部分：故事理解、概念关系抽取、剧本其它元素的抽取。具体内容如下：在故事理解中,本文选择了概念图这一知识表达工具对故事文本做出形式化描述。因为知识表达是本系统的基础,而概念图这种与自然语言互译的知识表示工具能够很好的进行语义分析和推理。本文改进了概念图的生成算法,提出了一种主线图的方法。主线图是在概念图生成的基础上,通过概念图的运算,得到包含主要角色和主要情节的图。通过主线图这一工具,获得对故事的总体理解和把握。概念关系的抽取主要是抽取童话故事中角色与场景、角色与角色以及角色与道具之间的关系。本文改进了传统的模板匹配方法,结合概念图的强大语义表达功能,通过总结常见关系和规则映射得到概念图模板库,进而利用概念图匹配运算来抽取相应关系。实验表明,采用基于概念图的方法,概念关系抽取的准确率和召回率得到有效的提高。剧本元素除了概念关系之外,还有角色名、道具名以及场景信息等。在此首先解决了未登录词的识别问题,提出了基于知网的混合识别方法进行识别角色与道具。由于在识别中增加了语义因素,这种方法识别的效果要优于传统的基于规则和统计的方法。针对场景信息中的时间地点等信息,通过建立规则库,采用基于规则的方式进行有效抽取。最后选取动画剧本标记语言CSML(cartoon scenario markup language)对抽取的信息进行描述,形成动画剧本。
[Abstract]:The rapid development of animation industry has led to economic growth and enriched people's lives, but the traditional animation production needs a lot of labor, animation automatic generation can effectively liberate human resources and save a lot of resources. Animation script generation is a key link in automatic animation generation, it is a bridge between the story text and the animation scene. How to extract the script elements needed for the animation script and choose the appropriate way to organize the script is the key to solve the problem. For the animation script generation research, there have been a considerable number of foreign research results, but due to the particularity of Chinese word-formation, there is still a lack of domestic research. This paper uses natural language understanding technology to divide the script generation system into three parts: story understanding, concept relation extraction, script other elements extraction. The details are as follows: In the process of story understanding, this paper chooses concept graph as a knowledge representation tool to formalize the story text. Because knowledge representation is the foundation of this system, concept map, which is a knowledge representation tool with natural language translation, can do semantic analysis and reasoning well. In this paper, the algorithm of generating concept graph is improved, and a method of principal graph is proposed. The main line graph is based on the generation of the concept graph, and through the operation of the concept graph, the graph containing the main characters and the main plot is obtained. Through the main line map this tool, obtains the overall understanding and the grasp to the story. The extraction of conceptual relations is mainly to extract the relationship between characters and scenes, roles and characters, and between characters and props in fairy tales. This paper improves the traditional template matching method, combines the powerful semantic expression function of the concept map, and obtains the concept map template library by summarizing the common relation and the rule mapping, and then extracts the corresponding relation by using the concept map matching operation. Experiments show that the accuracy and recall rate of concept relation extraction are improved effectively by using concept graph based method. In addition to conceptual relationships, script elements include role names, props names and scene information. In this paper, the problem of recognition of unrecorded words is first solved, and a hybrid recognition method based on knowledge net is proposed to identify characters and props. Due to the addition of semantic factors in recognition, this method is superior to the traditional rule-based and statistical methods. Aiming at the information of time and place in scene information, the rule base is established and the rule based approach is used to extract the information effectively. Finally, the animation script markup language CSML(cartoon scenario markup language) is selected to describe the extracted information to form the animation script.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.1

【参考文献】