Research And Application Of Question Answering Technologies In Traditional Chinese Medicine System

The Question Answering(QA)technology has been applied in the fields of financial insurance,e-commerce,etc,and the research of related methods has been a hot topic in recent years.However,there are still few Question Answering Systems in the Traditional Chinese Medicine area.Based on the Traditional Chinese Medicine System(TCM),a QA system is researched and constructed to meet the needs of its project construction.The main works are as follows:1.Construct a Question Answering database and datasets of TCM:the database has more than 1,500,000 pairs of questions and answers for providing retrieval services.The datasets include Similar Question Dataset(SMD)and Question-Answer Matching Dataset(QAMD),and the former is used to study the retrieval methods of similar questions,while the latter is used for question-answer pair(QA pair)based QA models.2.Research on retrieval methods of similar questions based on text representation:using Similar Question Dataset,we firstly studied the text representation method based on keywords and embedded representation.We proposed a joint representation method based on TFIDF,and showed the effectiveness of this method through experiments.Secondly,the similarity calculation methods of text representation were studied,and the calculation scheme of combining Sigmoid function and Manhattan distance was proposed,which made a good compromise between time and accuracy.3.Research on Q A models with Deep Learning:using Question Answer Matching Dataset,we introduced the BiMPM model,and proposed to extract matching features based on word embedding similarity and FM matching feature based on text representation.The validity of the proposed methods was verified by being compared with the QA-CNN model and other models.4.Based on the above research,the QA function was designed and realized in TCM with the methods having good effect.It uses a chat-box interface and supports message logging,which is convenient for users to use and facilitates the subsequent improvement.