Construction of Word Semantic Knowledge Graph in the Domain of Ancient Chinese Poetry

The knowledge in the field of ancient poetry lacks of structured summary. If knowledge graph can be used to connect the knowledge in this field, the intuitiveness and comprehensiveness of the study of ancient poetry will be greatly improved. Therefore, a construction method of the knowledge graph of ancient poetry is proposed. The candidate nodes of the knowledge graph of ancient poetry are obtained by means of the new word discovery method integrating multidimensional features, and then the relationship between nodes is established by using the semantic information in the HowNet language knowledge base. For the candidate nodes that are not marked in HowNet, the semantic prediction model is used to predict the semantic information. Through two experiments of word mining effect evaluation and meaning prediction, we verify the validity of the proposed knowledge graph construction method of ancient poetry. Furthermore, by testing the coverage of Chinese poetry graphs, we find that on the public CCPC dataset, the constructed knowledge graph has a coverage of 96.42%. This result shows that the knowledge map of ancient poetry constructed by us can provide a solid data support foundation for future downstream tasks in the field of ancient poetry.