Research On Chinese Health Question Classification Based On Deep Learning

With the rapid development of the information age,the Internet has increasingly become an important channel for people to obtain health information.Searching and consulting personal health issues through search engines and various convenient online inquiry platforms（such as Doctor Dingxiang,Xunyiwenyao,etc.）has gradually become the first step for people to ask for medical advice in hospitals.In the process of health consultation,the most crucial step is to accurately understand the intention category of the user’s question.For example,the intention category of the question"How to treat bone spurs in the cervical spine?"is to inquire about the"treatment plan",while the intention category of the question"How much does it cost for rhinoplasty?"is"medical expenses".Identifying the intent category of user queries can be treated as Chinese text classification problems.Chinese public health questions have the characteristics of scattered and diverse expressions and complexity.There is no unified format or expression style for public health consultation platforms,and the texts are varied in length and sparse in features,making it difficult for machine models to understand the intent or category of the problem.Current classification models only use the local topology structure of the text to construct the network,which to some extent ignores the impact of syntactic information on the classification results.Therefore,this article uses knowledge of deep learning to study the classification of Chinese health questions,and the main research results include:（1）Building a Chinese health question semantic extraction model that combines CBAM and multi-channel CNN models.To address the problem of varying lengths and sparse features in public health question texts,this paper proposes the classification method that combines CBAM and multi-channel CNN model.First,BERT pre-trained model is used to semantically encode the question,followed by multi-channel CNNs that extract text feature information using the Inception multi-branch idea.Then,to highlight the focus of visual attention on key features of the text,the model introduces CBAM attention mechanism to allocate weights to feature maps.Finally,the model classifies and outputs the results through a classification layer.Experimental results show that compared to the baseline network,the proposed method increases₁ value by 2.1%on Chinese health question dataset and has better classification performance compared to classical text classification models.（2）Building a syntactic analysis model based on graph neural networks.To extract syntax information between text,this paper designs a syntactic analysis model based on graph neural networks.Firstly,the model uses the BERT model to achieve semantic encoding of the text.Then,syntax mask matrices are constructed based on different syntax distances between words to enhance traditional GCN by combining adjacency matrices and syntax mask matrices.After that,syntactic features of the sentence-question pair are obtained through multiple graph convolution operations.Finally,the model achieves classification results through the classification layer.The model conducted multiple comparative experiments,and the results showed that the proposed syntactic analysis model based on graph neural networks has better classification performance and is more advantageous in processing Chinese health question dataset.（3）Building a Chinese health question classification model that combines multi-channel attention-based convolution and syntactic analysis.Firstly,this approach uses the BERT pre-trained language model to generate word vectors,alleviating the problem of semantic sparseness caused by short text.Then,it employs multi-layer convolution kernels of different scales to extract semantic features of different granularity to solve the problem of noise in public health questions.In addition,it integrates multi-scale feature attention mechanisms to suppress irrelevant semantic information and highlight deep text features.Meanwhile,through syntactic analysis,we construct syntactic dependency graphs and utilize graph attention networks to capture syntactic dependency relationships within the text.Finally,the extracted semantic features are combined with the syntactic structure,and the fused result is classified through a classification layer for question classification.Through model comparison experiments,the proposed model in this paper obtains 88.9%,89.1%and 89.0%in the P,R and₁ scores,respectively,which is significantly better than the current mainstream machine learning and deep learning models.This article addresses the problems in the existing Chinese health questions and proposes three models for Chinese health text classification using deep learning knowledge.Relevant experimental verifications are conducted,and the results show that the proposed models can effectively improve classification performance.