The Construction And Analysis Of The Knowledge Graph Of Mongolian Historical Figures

The preservation and utilization of knowledge is the source of the continuation of human civilization for thousands of years.In recent years,the development of machine learning and deep learning technology in the direction of natural language processing has made great progress in the research and application of knowledge graphs.The knowledge graph breaks the limitations of traditional data storage media,and integrates multi-source heterogeneous data on the Internet through entities and relationships in a structured manner to form a semantic network,which is expressed in a form closer to the human cognitive world,which is intelligent questions and answers,search,reasoning and recommendation provide the underlying data and technical support.Domain-specific knowledge graphs are usually used for deeper research in various fields due to the depth and completeness of their knowledge,the richness and rigor of data patterns,and the high accuracy of description.Inner Mongolia has a long history and culture,and there are also many books and historical materials that record these ethnic cultures.However,most of this knowledge is stored in the form of thesis text,which is very inconvenient to consult and learn.The knowledge graph can clearly display various types of knowledge,such as character relationships,historical events,and so on.Therefore,based on the investigation and analysis of existing general knowledge graphs,domain knowledge graphs,and key technologies of knowledge graph construction,this thesis uses Mongolian historical figures as the basis to identify named entities,relation extraction and classification in the construction of Mongolian domain knowledge graphs.And to carry out in-depth research on key technologies such as visual display.Focus on the following three aspects of research:(1)This thesis takes advantage of the BERT pre-training model and adopts domain-adaptive retraining methods to design a "pre-training + fine-tuning" language model H-BERT that meets the characteristics of the domain,and incorporates entity and vocabulary information features in the pre-training process。Multi-granular MASK technology conducts task-guided multi-task learning optimization training;and fine-tunes the model through tailored lightweight processing to achieve entity recognition and relationship classification.Achieved a leap from 0 to 1 in the knowledge map of Mongolian culture and history.(2)This thesis proposes entity recognition and relationship extraction methods,using the "H-BERT+" method for entity recognition and relationship extraction.Entity recognition models include basic model H-BERT+softmax,enhanced model H-BERT+CRF and H-BERT+Bi LSTM+CRF.The relationship classification model includes the basic model H-BERT+softmax and the enhanced model H-BERT+Bi LSTM+softmax.Three types of detailed experimental schemes are designed: evaluation for all(entity/relationship)categories;evaluation for sub-category(entity/relationship);case study for a single instance.From five measurement indicators such as accuracy,precision,recall,F value,and ROC curve,the effectiveness of the method proposed in this article is verified from different levels and dimensions.Experimental results show that this method has a high accuracy rate,and can correctly extract entity relationships and construct a knowledge graph of Mongolian historical figures.(3)This thesis studies the display of knowledge graphs and the application of knowledge graphs,builds a visualization system to show the relationship between Mongolian characters,and designs a simple question-and-answer system to query the relationship between the characters,and finally verifies the method proposed in this thesis.