【作者】 张娜；

【导师】 林怀忠；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2019， 硕士

【摘要】 文物是研究人类文明、历史、社会的重要物质“证据”。近年来,随着国家对于文物保护工作的关注,文物的宣传形式趋于多样化,人们对于文物的关注度明显上升,文物资源数字化需求增加。如何有效地组织和利用大量的文物数字资源是近些年来的关注热点。国内多是博物馆在进行专题展览时,会临时构建小规模的知识图谱来组织文物数据,这些文物知识图谱完全依赖于人工构建,缺乏自动化方法。本文基于以上背景,从实际应用出发,对文物知识图谱构建过程中的文物关系自动抽取技术进行了研究,设计并实现了完整的文物知识图谱构建与展示方案,具体工作如下:首先,本文提出了一种基于半监督学习的文物关系抽取算法,在算法中使用改进的Tri-training模型(三分类器协同训练模型),用于文物关系的自动抽取工作。其中包括:设计新的核函数来度量抽取结果与关系模式的相似程度,采用三种不同分类器构建改进的Tri-training模型。通过实验证明改进后的模型在文物关系的抽取上优于原Tri-training模型。然后,本文参考已有的知识图谱构建技术,设计并实现了面向文物领域的知识图谱构建方案。以书画、文献、名人故居为例,构建以黄宾虹为中心的文物知识图谱,通过文博专家对构建的文物知识图谱的评价和实际应用,说明了本文设计的文物知识图谱构建方案的可行性和准确性。最后,本文基于以上研究成果,设计并实现了一个文物知识图谱构建与展示系统。本系统支持构建各种专题文物知识图谱,支持文物知识结构调整、文物实例录入、文物文本数据处理、文物知识检索与动态知识图谱展示等功能。更多还原

【Abstract】 Cultural relics are important material "evidences" for studying human civilization.In recent years,with the attention of the state on the protection of cultural relics and diversified forms of publicity for cultural relics,people’s attention to cultural relics has increased significantly,and the demand for digital resources of cultural relics has increased.How to efficiently organize and utilize a large number of cultural relics’digital resources has become a research hotspot.In China,most museums will temporarily construct small-scale knowledge graphs to organize cultural relics’ data when conducting special exhibitions.These cultural relics knowledge graphs rely entirely on artificial construction and lack of automated methods.Based on the above background and from the perspective of practical application,this paper studies the automatic extraction technology of cultural relics’ relationship in the process of constructing cultural relics knowledge graph,and designs and realizes the construction and display scheme of cultural relics knowledge graph.The specific work is as follows:Firstly,this paper proposes a semi-supervised learning-based algorithm for extraction of cultural relics’,relations.In this algorithm,we use the improved tri-training model(collaborative training model with three classifiers)to extract cultural relics’ relations automatically.We design a new kernel function to measure the similarity between the extracted results and the relational model,and construct an improved Tri-training model using three different classifiers.It is proved by experiments that our improved model is superior to the original Tri-training model in the extraction of cultural relics’relations.Then,by referring to the existing knowledge mapping technology,this paper designs and implements a knowledge mapping scheme for the field of cultural relics.Taking calligraphy and painting,literature,and former residences of celebrities as examples,we constructed a cultural relics knowledge graph centered on Huang Binhong.Through the evaluation by experts and practical application of the cultural relics knowledge graph,we illustrate the feasibility and accuracy of the cultural relics knowledge graph construction scheme designed in this paper.Finally,based on the above research results,we design and implement a cultural relics knowledge graph construction and display system.This system supports the construction of various thematic cultural relics knowledge graphs and the cultural relics knowledge structure adjustment.This system also supports the functions of cultural relics’ data entry,cultural relics’ text data processing,cultural relics’knowledge retrieval and dynamic knowledge graph display.更多还原