基于图文数据的非遗跨模态知识图谱构建方法研究  

Research on the Construction Method of Intangible Cultural Heritage Cross-modal Knowledge Graph Based on Image and Text Data

作　　者：夏立新[1] 陈欢 李雪漫 翟姗姗[1,2] Xia Lixin;Chen Huan;Li Xueman;Zhai Shanshan(School of Information Management,Central China Normal University,Wuhan,430079;Intelligent Computing Laboratory for Cultural Heritage,Wuhan University,Hubei,430072)

机构地区：[1]华中师范大学信息管理学院,武汉430079 [2]武汉大学文化遗产智能计算实验室,湖北430072

出　　处：《情报资料工作》2024年第5期73-81,共9页Information and Documentation Services

基　　金：国家社会科学基金一般项目“数字人文视域下非遗知识图谱自动构建与长期演进研究”(批准号:20BTQ071);教育部哲学社会科学实验室-武汉大学文化遗产智能计算实验室开放基金项目“面向非遗领域的地方志数字资源语义标注与关联发现”(项目编号:2023ICLCH007)的研究成果之一。

摘　　要：[目的/意义]在多模态信息资源日益丰富的背景下,跨模态信息组织方式对非遗信息资源序化存储与共享利用具有重要价值。文章提出一种基于图文数据的非遗跨模态知识图谱构建方法。[方法/过程]文章通过结合多模态信息抽取与知识图谱技术对非遗数字资源进行知识重组。首先,从非遗项目内容与模态维度设计非遗跨模态知识图谱模式层,为图文数据组织提供顶层设计;其次,分别讨论面向非遗领域的图像与文本资源的知识抽取方案,获取图文模态下的非遗知识,并验证知识抽取方法的有效性;最后,通过面向不同场景下的一系列跨模态知识链接方法抽取非遗图文实体间的跨模态关系。[结果/结论]以楚剧非遗项目为例,通过多数据源、多技术结合的方式构建楚剧跨模态知识图谱,验证了文章所提出构建方法的可行性,对非遗数字化建设具有一定参考价值。[Purpose/significance]Under the background of increasingly rich multimodal information resources,cross modal information organization has important value for the sequential storage and sharing of intangible cultural heritage information resources.This paper proposes a method of constructing intangible cultural heritage cross modal knowledge graph for image and text data.[Method/process]This paper combined multimodal information extraction and knowledge graph technology to reorganize the knowledge of intangible cultural heritage digital resources.Firstly,the schema layer of intangible cultural heritage cross modal knowledge graph is designed from the content and modal dimensions of intangible cultural heritage digital resources to provide top-level design for graphic data organization.Secondly,this paper discusses the knowledge extraction scheme of image and text resources in the field of intangible cultural heritage,obtains the intangible cultural heritage knowledge in the mode of image and text,and verifies the effectiveness of the knowledge extraction method.Finally,a series of cross modal knowledge linking methods for different scenes are used to extract the cross-modal relationships between intangible cultural heritage graphic entities.[Result/conclusion]Taking the intangible cultural heritage project of Chu opera as an example,the cross-modal knowledge graph of Chu opera was constructed through the combination of multiple data sources and technologies,which verified the feasibility of the construction method proposed in this study,and had a certain reference value for the digital construction of intangible cultural heritage.

关 键 词：非物质文化遗产 跨模态 知识图谱 图文数据 知识抽取 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]