| In recent years,artificial intelligence has developed rapidly,and question answering systems are being applied to more and more scenarios.However,the Chinese language has a complex structure and flexible parts of speech.The answers output by question answering systems in specific fields often cannot meet the needs of users.With the development of neural networks,under the training of large-scale professional background knowledge,question matching can be combined with semantics,and the accuracy of semantic discrimination of questions in the Chinese context has been improved.However,the large-scale professional domain corpus and the accuracy of the question matching algorithm are still key issues in the research of question answering systems.Under the above background,this article mainly carried out the following work:(1)Constructed a data set for a specific field.This paper constructs a structured data set based on the field of social security.According to the template questions,extended questions and standard answers provided by professionals,the collected data will be classified and labelled,taking into account the imbalance problems in the subsequent question matching algorithm.Research,control the ratio of positive and negative samples when labeling,and make it adjustable as needed.(2)The stability evaluation of the question-answer matching model based on confidence is studied.Based on the statistical confidence theory,this paper designs a method for evaluating the stability of the model’s problem matching accuracy.It adopts a two-way full self-attention network(BERT)problem matching technology.Aiming at the problem of data imbalance,a focal loss function is proposed.(Focal-loss)instead of the original loss,use multi-model fusion(K-fold)to prevent the model from over-fitting and improve its robustness.Finally,the improved model and baseline are tested for stability to ensure the quality of system question matching.(3)Designed and implemented a question and answer system.According to the different functions of the question processing,document processing,and answer processing modules of the question answering system,this paper designs and implements a question answering system in the social security field,and tests the system performance. |