| In the context of the booming Internet and accelerated digitalization,people can acquire all kinds of knowledge through the Internet,but how to efficiently filter out valuable information from massive data has become an urgent problem to be solved.Question-answering(QA)systems,through semantic analysis and other techniques,can deeply understand the users’ search intentions and provide fast and accurate answers to meet the user’s needs for precise positioning of answers.This paper investigates the question classification and answer selection tasks of the FAQ system,and uses deep learning techniques to improve the intelligence of QA system in Chinese context.To address the issue of difficulty in extracting and representing semantic features in question classification task,this paper proposes the CW-Bi GRU-MA model.The model uses word and character dual-granularity embedding inputs,combines Bi GRU for feature extraction to enrich the semantic features of the question text,and introduces a multi-head attention mechanism to strengthen the expression of important features,improving the question classification performance.To address the problem of inadequate semantic matching between questions and answers in answer selection task,this paper proposes the BERT-MPAM model,using a comparative aggregation network structure to avoid missing semantic information to a greater extent,and using BERT for word embedding to overcome the limitations of traditional word vector techniques.The model adopts a multi-perspective attention mechanism to comprehensively extract the local and interaction semantic features of the question and answer,and effectively aggregate these two attention networks,improving the matching performance of the model.This paper conducts comparative experiments of the CW-Bi GRU-MA model and the BERT-MPAM model on their respective public datasets,verifying that they have good performance.Finally,this paper integrates the above two models and constructs a semantic matching-based financial QA system,conducting a detailed analysis and design of the system.The system can deeply analyze users’ questions,query the QA library,and return answers to help people quickly and accurately obtain financial knowledge,and has practicality. |
