Knowledge graph-based intelligent Q&A system for Korean culture with hybrid reasoning mechanism

This paper studies the design and reasoning mechanism of Korean cultural intelligent question answering system (QAS) based on Knowledge graph (KG). In view of the limitations of the existing cultural communication methods in providing systematic knowledge, this study constructs a structured Korean cultural knowledge map, which adopts a directed acyclic graph (DAG) structure, covering many cultural types such as history, art and customs, and fuses multi-source heterogeneous data through an improved TF-PDF algorithm to solve the problem of Korean homographs. In the stage of knowledge extraction and representation, a hybrid model combining BERT-ETRI, Conditional random field (CRF) and symbolic rule base is proposed to improve the recognition accuracy of culture-specific expressions, and an improved TransE model is used for knowledge embedding. As for the reasoning mechanism, a multi-hop reasoning mechanism combining symbolic logic and deep learning is proposed, and a framework combining probabilistic logic programming with graph neural network (GNN) is adopted, giving consideration to interpretability and generalization ability. The experiment adopts QA-3000 test set, which contains 3000 questions about Korean culture. The results show that the research system is superior to the baseline method in cultural accuracy (CA), F1 value and logical coherence (LC), and the response time (RT) is equivalent to that of the general KG system. In addition, a cultural concept drift detection model is designed to verify the influence of different update cycles on the dynamic update effect of KG, which shows that more frequent updates can significantly improve the accuracy and adaptability of the system.