Research And Implementation Of FAQ Question Answering System Based On Deep Learning

With the rapid development of science and technology,the data on the internet is growing exponentially.Although the abundance of data enhances the possibility of users acquiring knowledge from multiple aspects,it also creates the problem of information overload where users cannot locate valid information.To solve this problem,intelligent questionanswering systems have emerged.These systems use intent recognition and semantic understanding to provide users with a concise natural language answer corresponding to their question,saving users time and providing accurate answers.This thesis focuses on the key issues of FAQ question-answering based on deep learning,including the automatic generation and expansion of question-answer(Q&A)pairs based on unstructured text,simplification of complex sentences,and question matching strategies.For the automatic generation and expansion of Q&A pairs,the BNCVAE-QAG model based on conditional variational autoencoder is proposed.The model uses batch normalization layers to prevent the Kullback-Leibler divergence from disappearing.The model also extracts spatiotemporal features in the text through a combination of Bi-LSTM and CNN,capturing the inherent connections in the text.Self-attention mechanism is used to obtain the relationship between the content and the answer.A question generalization mechanism is proposed to generate more Q&A pairs.For the simplification of complex sentences,the seq2seq_mt5 model based on the seq2seq structure is proposed.The model uses self-attention mechanism in the encoding layer to capture the long-distance dependencies in the sentence,and self-attention mechanism and cross-attention mechanism in the decoding layer to obtain joint features of the text.For question matching,the Qq+Qa+Qc matching strategy is proposed,where Qq refers to the semantic similarity between the user’s question and the questions in the question-answer set,Qa refers to the potential correlation between the user’s question and the answer in the question-answer set,and Qc refers to the relevance between the user’s question and the original corpus.This strategy combines Qq+Qa to improve the accuracy of intelligent questionanswering systems and introduces the Qc strategy to expand the search range and give users more choices.Based on the above technologies,the design and implementation of an intelligent question-answering prototype system are completed.The experimental results show that the proposed BNCVAE-QAG model has an EM of 80.64%and F1 of 87.72%.The proposed complex sentence simplification model achieves a Rouge-1 of 68.89%,Rouge-2 of 55.24%,and Rouge-L of 66.89%.In complex situations,the proposed Qq+Qa+Qc question matching strategy can also achieve an accuracy of 88%.The deep learning-based FAQ question-answering system designed and implemented in this article provides users with complete questionanswering functionality,meets users’ consulting needs,and reduces the cost of manual customer service.