Research On Chinese Healthy Question Classification Based On Deep Learning

With the development of information technology and the continuous improvement of people's living standards,medical health has become a hot topic of public concern.Exploiting search engines,question answering systems,etc.to consult medical health problems has gradually become the mainstream way of public health consultation.The question answering system,as an online platform with high efficiency and interactive humanization,reduces the difficulty of using online users and eases the consultation pressure of health consulting experts.During the interaction process,the question answering system must understand the intention of the user's question accurrately.Only when the user's question is understood correctly,question answering system can return the expected answer to the user accurately.For the classification of health problems in the online user consultation process,this article has carried out the following research:(1)A Rough classification of health questions combining local semantics and global structure.Given the shortcomings of existing health question classification methods,this paper proposes a health question classification method combining local semantics and global structure.First,the local semantic representation and global structural representation of sentences are obtained through convolutional neural networks and independent recurrent neural networks,respectively.Then,the self-attention mechanism is used to fuse the local semantic representation vector and the global structural representation vector to obtain the final semantic representation vector of the sentence.Finally,the classification layer is used to classify and output the classification results.Experimental results show that the method of this paper improves the problem of gradient disappearance and gradient explosion effectively.Compared with existing methods,this method achieves a higher F score on the Chinese health question dataset.(2)A detailed classification of health questions combined with external knowledge and double-layer attention mechanism.Aiming at the problem of online users' non-standard wording,this paper proposes a deep neural network classification method based on external knowledge and double-layer attention mechanism.First,the word isencoded through a bidirectional GRU;then,the candidate word of the current word is obtained through the vocabulary,and the vector representation of the current word is obtained through the first-level attention mechanism;then,the health question's final semantic representation is obtained through the second-level attention mechanism;finally we classify through the classification layer and output the result.Experimental results show that the method in this paper enriches the information of word meaning,alleviates the problem of non-standard words used by consumers and improves the classification effect of a small amount of text.(3)Multi-label classification of health questions based on codec structure.Aiming at the multi-topic problems raised by online users,this paper proposes a multi-label classification method based on convolutional neural network,long short-term memory network and attention mechanism.First,the convolutional neural network is used to obtain continuous local semantic features;then,the long-term and short-term memory network is used to model the sequence to obtain the initial encoded representation of the healthy question;then,we use the attention mechanism to obtain more attention to generate the final encoded representation of the health question;finally,the tag sequence is output through the decoder.Experimental results show that the method in this paper not only establishes the connection between tags,but also makes the connection between tags and sentences closer.