Constructing a semantic-driven QA system for TCM cultural archives via deep learning and knowledge graph fusion

As an important carrier of cultural communication, the museum of traditional Chinese medicine (TCM) has some problems, such as information fragmentation and low retrieval efficiency, which cannot meet the needs of modern users. The purpose of this study is to construct the Intelligent Question Answering System (IQAS) of TCM Museum based on knowledge graph (KG), so as to improve the interactive experience of users. By integrating museum collections, ancient books and clinical data, TCMKG is constructed, and a semantic algorithm that can identify complex terms and metaphorical expressions in TCM field is designed. In the aspect of KG construction, the six-tuple model is used for ontology modeling, the improved BiLSTM-CRF model is used for named entity identification, and the relationship extraction is based on the Attention-BiGRU model, and the knowledge fusion and storage are realized by using the double-layer mixed storage mode. In IQAS design, the intention classification system is constructed, and the mixed model of BERT and CNN is used to classify intentions, and the multi-hop reasoning algorithm based on reinforcement learning is introduced to deal with complex query problems. The system also designed a metaphor processing model to map non-standard terms to standard diseases. The test and evaluation results show that the system performs well in entity recognition, simple query, multi-hop reasoning and metaphor processing, in which the F1 value of entity recognition is up to 0.93, the accuracy of simple query is 94.3%, the accuracy of multi-hop reasoning is 82.7%, and the accuracy of metaphor processing is 78.5%. The system not only improves the intelligent level of knowledge service in TCM Museum, but also provides new technical support for the inheritance and innovation of TCM culture.