Research On The Intelligent Question-answering System Of Treatise On Febrile Diseases Based On Knowledge Grap

Objective:Following the provision of a Chinese solution to the COVID-19 by Traditional Chinese Medicine(TCM),public recognition of TCM increases the demand for quick and accurate access to TCM knowledge,while also posing new challenges to TCM informatization.Using the Treatise on Febrile Diseases of the Song Dynasty as an example,the paper aims to lay the groundwork for the informationization of ancient Chinese medicine books by combining knowledge graph(KG)and intelligent question and answer(QA)technology.Methods:(1)The rule-based method was used to extract the five TCM entities of disease,syndrome,symptom,prescription,and medicine,as well as their relationships and attributes,from Treatise on Febrile Diseases,and KG was managed by attribute graph.Finally,Neo4 j was used to store KG.(2)NER training corpus consists of five types of TCM entities extracted using the rule method.The BiLSTM model was used for feature extraction,and the CRF model was used for output annotation.The Embedding layer experiments with word2 vec,BERT,and ALBERT models and chooses the model that is best suited for entityextraction in Treatise on Febrile Diseases for NER based on user input questions.(3)The different expressions of five types of common questions in Treatiseon Febrile Diseases were collected to create a Relational Classification(RC)training dataset.The training of a multinomial Naive Bayes(Multinomial NB)model yielded the RC model of the QA system of Treatise on Febrile Diseases.(4)Cypher query templates are designed for various types of questions and are usedto query and return answers to questions raised by users in Neo4 j.Results:(1)The Treatise on Febrile Diseases KG,which contained 639 entities and2076 entity relationships,was created,and KG was managed and stored using attribute graph and the Neo4 j.(2)The accuracy rate,recall rate,and F1-measure values of Embedding using the ALBERT model are 85.37%,86.84%,and 86.02%,respectively,which are higher than those using the word2 vec and BERT models.It is better suited for NER of user-input questions in Treatise on Febrile Diseases QA system.(3)In data set classification,the accuracy rate,recall rate,and F1-score value of multinomial NB model are 92.00%,92.00%,and 91.00%,respectively,and the classification effect is good,which can be applied to QA system of Treatise on Febrile Diseases.Conclusion: In the paper,basic knowledge of Treatise on Febrile Diseases can be quickly obtained by extracting KG based on entity,relationship,and attribute of Treatise on Febrile Diseases and combining it with the related technology of intelligent QA,which lays a foundation for the informationization of ancient Chinese medicine books.