Research On Intelligent Question Answering Model Based On Deep Learning

When retrieving colloquial texts or professional terms,traditional search engines can no longer meet actual needs.With the wide application of deep learning technology,intelligent question and answer can be more suitable for different application fields.Intelligent question answering is a process of retrieving knowledge information with a high degree of matching in the question database through the understanding of the deep semantic information of the text,and giving feedback on the reply sentences.Facing the shortcomings of the existing intelligent question answering model,this paper makes improvements in named entity recognition,text classification and text matching tasks,so as to improve the accuracy of intelligent question answering results.The specific research content is as follows:First of all,in view of the existing named entity model facing the problem of unclear division of complex entity vocabulary boundaries,this paper proposes a named entity model based on deep learning,which incorporates a vocabulary enhancement algorithm based on the WoBert model,so that the model can introduce external vocabulary information,To enhance the model's understanding of vocabulary features,thereby effectively dividing entity vocabulary boundaries and entity categories.Experiments were conducted on the public data set,and the harmonic average F1 of precision rate and recall rate was selected as the model evaluation index.The results showed that the F1 value of this model reached 82.19%,which was 7%higher than the effect of the Bert-CRF model.Secondly,when faced with complex long text sentences,it is difficult for a single-layer neural network based on deep learning to understand the real semantic information of the text.Therefore,this paper proposes a multi-feature extraction fusion network model based on the self-attention mechanism,which uses the self-attention mechanism to understand the text The dependent features of the sentence in the context structure,using the deep convolutional attention overlay network and the attention mechanism-based two-way gated combination network to deeply mine the text semantics,and then stitch the two extracted semantic features into a fusion feature representation Vector,through the use of Softmax classifier to achieve classification of text features.After comparative experiments on public data sets,the results show that the F1 value of this model is 11.6%higher than that of the CNN model.Then,aiming at the poor effect of existing text matching models,a text matching model using multi-layer and multi-feature extraction and fusion is proposed.The text matching model is divided into two layers,one layer is the text matching model based on the twin Bert model multi-feature extraction and fusion,and the other layer is the similarity algorithm of multi-feature fusion.Realize the layer-by-layer screening of the text of the question and answer library,and in-depth understanding of the true semantics of the text,thereby improving the text matching effect.The text matching model is tested experimentally on the public data set.The F1 value of the model in this paper is as high as 70.75%.The experimental test of the similarity algorithm of multi-feature fusion shows that the text similarity calculation of multi-feature fusion is more in line with user needs.Finally,the intelligent question-and-answer model is applied to the Xunyiwenyao intelligent question-and-answer system to achieve real information interaction with users,and to effectively understand professional domain knowledge.After multiple sets of tests,the accuracy of the question and answer reaches 74%,which verifies that The feasibility and effectiveness of the model in this paper.