Research On Construction Technology Of Knowledge Graph Of Cultural Relics Collection

Cultural relics are the witnesses of China’s long history and culture,and museums have always been the carriers of preservation and display of cultural relics.With the rapid development of artificial intelligence,intelligent construction has become the focus of museums.However,the current management of cultural relics information in museums has the following two shortcomings: The information of cultural relics in a single museum is scattered and independent,lacking relevance;It is difficult for different museums to communicate with each other due to different preservation strategies of cultural relic information.It has caused a huge obstacle to the intelligent construction of museums.Knowledge atlas of cultural relics collection organizes cultural relics information into triplet form by excavating the deep relationship between various entities related to cultural relics,so as to connect cultural relics and museums,laying a foundation for the intelligent construction of museums.Starting from the intelligent construction of the museum,this thesis carries out the research and realization of the construction technology of the knowledge atlas of cultural relics collection,and mainly completes the following work:(1)Construction and design of knowledge atlas of cultural relics collection.By using the top-down construction method of knowledge atlas of cultural relics collection,five modules of knowledge atlas construction of cultural relics collection are determined: data source crawling module,cultural relics collection ontology construction module,cultural relics collection knowledge extraction module,cultural relics collection knowledge fusion module,cultural relics collection knowledge storage module.(2)Knowledge extraction module involves named entity recognition task.Aiming at the defect that Bi LSTM cannot remember all historical information,b TBL-CRF model is proposed in this thesis,and experiments are carried out on the constructed data set of named entity recognition of cultural relics collection.The results show that compared with BILSTM-CRF model,the F1 value of this method improves 2.14% in the cultural relic collection data set.(3)Knowledge extraction module involves relational extraction task.Aiming at the deficiency of insufficient information in traditional word vector,Word2 Vec and Trans E are combined to train word vector for remote supervised relationship extraction model with collection-related triplet information.This method integrates more relationship information between entities.Finally,this thesis conducted experiments on the cultural relics collection data set constructed by ourselves.The results showed that,compared with the word vector trained by Word2 Vec alone,F1 improved by 0.6% and AUC improved by 1%.(4)Realize the knowledge map of cultural relics collection.First crawl relics related information and data cleaning,and then determine the collection of cultural relics ontology in four categories,22,nine object attribute data,then use the proposed two models complete knowledge extraction,then use the baidu search page text similarity way entity disambiguation,using baidu encyclopedia alias with synonyms completed refers to the digestion,Attribute alignment was completed by comparing text similarity based on semantics.Finally,Neo4 j was used to store knowledge maps of cultural relics in The National Museum of China.