Research And Implementation Of Question Classification Method In Question Answering System

With the continuous progress of the Internet,the text on the network is exploding,in this information contains a lot of information,how to quickly get the information you want has become the focus.People often use search engines to search for the information they want,but the amount of information increases with the return of relevant pages,especially in specific areas such as law,medical care,etc.,it is difficult to meet the user's needs for critical information.Question answering system(QA)provides users with the information they need quickly and accurately,giving accurate answers accordingly when given a question.Question answering system consists of three parts: question analysis,information retrieval and answer extraction.Question classification is the initial part of QA,and whether it can correctly classify questions will directly affect the subsequent answer extraction.Question classification plays an important role in question answering system mainly in two aspects: on the one hand,assigning labels to questions according to the expected answer types,which is the basis of question classification,so as to narrow the range of candidate answers;On the other hand,for different types of questions,the question answering system will formulate different strategies in the subsequent operation.There are three problems in the current research of question classification: 1.It needs data with high relevance as the basis,and expresses the problem according to the data set of question classification,the feature of manual extraction,or combination of some features.Therefore,it has strong subjectivity and diversified language expression,This means that it is expensive to make accurate feature extraction methods by hand;2.Some of the classification models of question sentences are relatively simple in application,and the model itself has some defects,which can not fully learn the feature information and semantic information of the question sentences;3.Compared with the open data set of English question classification,there are less open corpus in Chinese,which is one of the main reasons that restrict the development of the research on Chinese question classification.In view of the above problems,the main contents of this paper are as follows:(1)For the current situation that Chinese open question answering corpus is relatively scarce,this paper uses the extension method based on the named aspect recognition technology and the replacement of interrogative words to transform Chinese declarative sentences into interrogative sentences,which can effectively transform simple declarative texts into interrogative sentences.In addition,we collected and constructed a certain scale of Chinese question and answering corpus including answer information from the question and answering platforms such as Sogou and Baidu know.(2)In order to better understand the users' intention of questioning and capture the deep feature information of question sentences,we choose to use the question classification model based on Transformer Encoder and Bi GRU and Attention mechanism.The results of the experiments show that the method is effective.(3)In view of the complexity and diversity of questions,construction of question classification system based on the above research,which integrates the comparative experiment and the deep neural network model proposed in this paper.Django is used to develop the background of web application,HTML5 + CSS3 is used to make the foreground page,and Docker is used to package the service environment.Through the performance test,the system has realized the function of question classification and has good concurrent access ability.