Research On Chinese Medical Question Answering Technology Based On Attention Mechanism

With the gradual formation of the global village,the Internet has begun to enter every household.Compared with people ten years ago,people now can access the Internet more conveniently,and offline medical resources are in short supply.People will give more priority to using the Internet to go online for consulting their own physical problems.This problem was considered and the state and hospitals solved it and hired relevant engineers to develop an online medical consultation platform.However,as more and more people are online,there are more and more consultation questions and there are not enough doctors to answer the questions in the background.Therefore,this paper consider how to use the existing technology to make the background automatically answer the patient’s questions.This is important problems.This paper consider that most patients have similar questions and the answers to similar questions must be similar.So we consider to use question answer matching methods to solve them.Because the medical field is different from the general field,this paper has carried out research on this.So the model can’t be just applied to the Chinese medical question and answer matching scene.There are two key issues that need to be solved: 1.For the text that cannot be directly processed by the computer,this paper need to consider how to convert into the best handleable numerical form,which can be also applied to the medical field.2.How to build a neural network to better capture the semantic features of medical text after obtaining the text in numerical form.For the first question,this paper compare the two types of fine-grained words and characters to process sentences in the text,and then encode the word sequence and word sequence.This article found that due to the specialty of the Chinese medical field.The current word segmentation tools are very poorly effective,which leads to word segmentation errors that will affect the follow-up tasks.And when the model use word embedding,it does not require word segmentation,so the model can avoid the problem of word segmentation errors.And this paper consider to instead of word embedding to encode the text.The two experiments found that the word embedding effect is better than the word embedding.So the conclusion of the experimental is that the word embedding is adopted,which can reduce the dimension of the matrix representing the text,and reduce the storage occupation and the number of calculations.The effect of the overall task can be improved.For the second question,firstly,on the two data of c Med QA and c Med QA2.0,this paper verified the traditional recurrent neural network and the different deformations of the convolutional neural network to do the matching effect of Chinese medical question answering.And this paper compare and analyze them for different models inferior.After introducing the attention mechanism,we obtain a stack convolutional neural network based on the attention mechanism.And this paper compare it with the unchanged model.The conclusion is that the effect of the task can be improved when the attention mechanism is introduced.In order to better complete the matching task of Chinese medical question answering,this paper consider further improving the effect of the model and improving the robustness of the model to different data sets.This paper also do experiments on the c MQA dataset.The c MQA dataset is for others to use crawler technology.The question and answer data of many different Chinese medical consultation platforms have been collected.And the proportion of negative samples has increased.In the professional field discussed in this article,this dataset is currently the largest dataset compared to other datasets.This paper verifies the above model on the c MQA dataset.Then this paper optimize the stack convolutional neural network that introduces the attention mechanism.And this paper add a multi-scale convolution kernel to the original model,which can increase the semantics that can be captured in breadth.The optimized model was verified.Experiments were carried out on the above several datasets,and several modules were compared,which proved that the introduction of attention mechanism into each optimized module.And the introduction of attention mechanism into the improved model can improve the effect.For the construction of the Chinese medical question answering system,this article provides a reference algorithm,the effect of which can meet the demand.