| With the development of artificial intelligence technology,knowledge graph has shown extraordinary results in many fields such as finance and medical care,and knowledge graph in the cultural field is also playing an increasingly important role.In recent years,knowledge graphs in the field of culture at home and abroad have appeared one after another.However,there is still a lack of research on knowledge graph in the field of Chinese mythology.Therefore,building a knowledge graph of Chinese mythology is helpful for the development of the cultural industry,and has considerable academic value and practical significance.As the key step in building a knowledge graph,the effect of entity relation extraction directly affects the quality of the knowledge graph.In addition,the entity popularity in the knowledge graph is also a significant feature for downstream tasks.In recent years,the researches on entity relation extraction have mainly focused on the mining of text features,however,the influence of the location interval between entities has been ignored.To address these problems,an interval embedding based two-stage entity relation joint extraction approach is proposed in this thesis,which models the relative interval of head and tail entities.The approach mines interval features through learnable embeddings,and embeds interval features into the two-stage entity relation joint extraction model.Due to the systematic exposure bias problem in two-stage models,a scheduled sampling based two-stage approach is proposed.The scheduled sampling of training dataset generates training data that approximates the predicted scenario.The loss function is optimized in the thesis,and the consistency of data distribution between training and prediction scenarios are strengthened.Finally,exposure bias is minimized.A content attention mechanism based entity popularity prediction method is proposed in this thesis.The entity features and the popularity feature contained in the relationship are deeply integrated by content attention mechanism,and Poisson regression is employed to predict the popularity.Experiments results show that the approach proposed in this thesis can perform better results compared with existing methods.In this thesis,Chinese Mythology Knowledge Graph(CMKG)is designed and implemented.CMKG includes main modules of intelligent retrieval,intelligent QA,creation assistant and visualization.Intelligent retrieval and QA help users quickly and easily acquire knowledge in the knowledge graph.Creation assistant provides technical support for the secondary creation of Chinese mythology,and the visualization module visually displays the knowledge graph of Chinese mythology.Finally,systematic tests on the CMKG are conducted,and the result shows that the functions of CMKG are complete with high usability and robustness,and CMKG provides considerable support for research and creation in the cultural field. |
