Research On Question-Answering System Related Techniques And Their Application In Traditional Chinese Medicine Field

The rapid development and wide spread of the Internet has created an era of data explosion for us.In the era of big data,the acquisition of information has become convenient and fast,but it is difficult to screen effective information from massive data.Question-answering system is an advanced form of information retrieval.Compared with the traditional keyword based retrieval method,the question-answering system can directly read the questions raised in natural language,understand the intention of the questions,and help people get concise and accurate answers.Question-answering based on knowledge base and machine reading comprehension are two hotspots in the research of question-answering system.The former takes a structured knowledge base as the source of answers,while the latter is mainly responsible for mining answers from relevant paragraphs.From the perspective of methods,the research hotspots of QA system mainly focus on two aspects: text-based features and neural networks.In the question-ansering method based on text features,it is difficult to model the question-answering system by combining various text features.In the question-answering method based on neural network,how to train the neural network model better has always been a concern.Traditional Chinese medicine(TCM)is the treasure of the Chinese nation.It records the experience and knowledge of the Chinese people in fighting against diseases and pursuing health for thousands of years.Since the 19 th century,the medical system of modern western countries has entered China,which makes TCM face great challenges.The theory of TCM was formed at an early stage,and a large number of its knowledge theories were recorded in ancient texts,which made it difficult to acquire knowledge of TCM.The question-answering system in the field of TCM can not only serve researchers of TCM,but also make it more convenient for ordinary users to understand the knowledge of TCM.Therefore,the research on question-answering system in the field of TCM is of great significance.In this paper,question-answering based on knowledge base and machine reading comprehension are studied,and these two kinds of question-answering system technologies are combined and applied in the field of TCM,and a question-answering system in the field of TCM is designed and implemented.Specifically,the main work of this paper can be summarized as follows:(1)This thesis proposes a knowledge base question-answering method based on multi-level text features.The QA system mines candidate triples from the knowledge base,At the level of sentence,word and character,multiple text features of the problem are extracted and their confidence scores are calculated.Experiments are carried out on multiple data sets and the experimental results are analyzed from multiple perspectives to verify the effectiveness of the proposed method(2)In this thesis,a joint training method for recurrent neural networks is proposed.This method introduces the auxiliary loss function of the shallow recurrent neural network and the deep recurrent neural network to update the model parameters.Experimental results on multiple datasets demonstrate the effectiveness of the proposed method.Compared with the traditional single-loss function training method,for the neural network model with deeper layers,the combined training method proposed in this thesis can achieve greater effect improvement.(3)This thesis proposes a word vector training method combining text features.This method combines the part of speech features and named entity categories of the text,cocodes the unknown words and trains their word vectors.Experimental results on multiple datasets demonstrate the effectiveness of the proposed method.The experimental results obtained by this method gain quite more improvements comparing with those obtained by other methods,when the test set with a higher rate of unknown words.In addition,the method proposed in this thesis takes up less resources and has higher efficiency when the same experimental measure is achieved.(4)This thesis proposes a question-answering system in the field of TCM.In cooperation with TCM experts,we designed TCM ontology structure and constructed TCM domain knowledge base based on this structure.The question-answering system proposed in this paper uses data from a variety of TCM fields as the source of answers,including the knowledge base of TCM fields constructed and unstructured documents of TCM fields containing medical cases.The question-answering system combined with multiple knowledge sources proposed in this thesis can achieve better performance than the traditional question-answering system based only on knowledge base.