Research And Implementation Of BERT-based Insurance Domain Question Answering System

With the in-depth study of artificial intelligence technology,the general field question answering system has been widely accepted.However,due to the small amount of text in the vertical domain,text representation hinders the development of vertical field question and answer systems.In order to make the vertical field question answering system more efficient and accurate,this paper proposes a smart question answering method based on BERT model for the insurance field.Compared with the traditional method,the QACNN model of the insurance corpus author and the improved QALSTM model,it proves that the BERT model is effectiveness in the insurance field.Finally,using the insurance domain question and answer corpus,complete the answer selection task,and then use the deep learning model BERT based method and migration learning strategy to construct the vertical domain question answering system.The main work of this topic includes:(1)We analyzed the background meaning and current situation of intelligent question and answer,and on this basis,studied the basic theories and techniques involved in intelligent question and answer.(2)We conducted algorithm research in the field of insurance based on the BERT model.First,we performed preprocessing operations on the insurance field data,including the conversion of the data format,the segmentation of the data,and the removal of stop words from the data.Then we implemented the traditional method and the intelligent question answering of the BERT model in the insurance field.In addition,we also compared the impact of using different similarity calculation methods on intelligent question answering in the insurance field.(3)We implemented a question and answer system in the insurance field.By analyzing the effects of the traditional method,BERT model,QACNN model and QALSTM model in the experiment,the BERT model and cosine distance were selected as the similarity calculation method to build the system.On win10,Pycharm was used as a development tool,and a set of insurance question and answer systemwas independently developed using python.The system mainly includes: question analysis,answer selection,feedback statistics,and user-related functions.Finally,users' feedback was used to solve future optimization problems.