A Research Of Text Retrieval Technology In Question Answering System

With the development of the internet and digital technology,there are numerous data transmitting on the network and it is difficult for traditional teams to meet the various real-time data needs of users.Therefore,the intelligent question answering system launched,and it has attracted the attention of many researchers in recent years.Text retrieval is a key component of the intelligent question answering system.Its main task is to search for similar texts.How to perform high-precision semantic matching and similar text retrieval are the key problems.Through the investigation of text retrieval technology in question answering system,this thesis analyzes the shortcomings of existing work,and proposes corresponding improvement methods.The main contributions of the thesis are as follows:1.In terms of semantic matching,the current methods are usually modeled based on the current input sample,and the information mining of other dimensions is insufficient,resulting in inaccurate semantic matching in some scenarios.Aiming at the above problems,this thesis proposes a multidimensional information-fused semantic matching model.In the information extraction layer,the model mines the information features related to text pairs from the word dimension,semantic dimension,and knowledge dimension.In the feature fusion layer,a comprehensive matching similarity is calculated by designing an appropriate feature fusion method.The experimental results verify the effectiveness of the information features mined by the model in this thesis.The results show that the model achieves better performance than the current mainstream models,and compared with the baseline BERT model,the F1 value increased by about 4%.2.In terms of similar text retrieval,traditional methods in the retrieval stage usually lack semantic information,and the information dimension is considered insufficient in the sorting stage.To improve the above problems,this thesis proposes a multidimensional information-fused similar text retrieval algorithm.The algorithm combines the sparse vector space model and dense retrieval in the retrieval stage,and uses the multidimensional information-fused semantic matching model proposed in this thesis in the sorting stage.The experimental results verify that better performance in the retrieval or sorting stage will improve the accuracy of the whole text retrieval.The results show that the performance of this algorithm is significantly better than the baseline method,compared with the baseline method BM25 + BERT,the P@1 index increased by about7%.3.Designs and implements a FAQ question answering system for banking business.Based on the similar text retrieval algorithm proposed in this thesis,the system is built using the architecture of distributed microservices.It can perform man-machine dialogue and knowledge base management.The usability of the core functions of the system is verified through the interface test,and the reliability through the stress test.Finally,the front-end page shows that the system can answer users' questions in real-time and accurately,and can manage the data of the knowledge base concisely and conveniently.This thesis proposes a multidimensional information-fused semantic matching model and a multidimensional information-fused similar text retrieval algorithm,which achieves better performance than the baseline methods.On this basis,a FAQ question answering system for banking business is implemented,and the core functions and stability of the system are tested.