A Study On Biomedical Text Semantic Matching Based On Deep Neural Etwork

In recent years,the amount of data generated in the biomedical field has grown exponentially.With the improvement of living standards,people are increasingly concerned about their health.Due to the continuous iteration and update of modern technology,more and more people tend to search for medical information and obtain health assistance on the Internet.However,traditional search engines have many shortcomings.Sometimes,they return a large number of repetitive or irrelevant content for medical queries that are phrased in a colloquial manner.Even medical professionals need to spend a lot of time to identify the accuracy of information in the vast and complex web pages.Most users lack relevant domain knowledge,making it difficult to accurately search for useful information in the returned results.This paper mainly explores the inability of current question-answering matching models to fully extract useful medical information from text,resulting in inefficient matching results.By leveraging current computer technology,this paper aims to address the related medical resource shortage issues in Chinese society.To solve the text matching problem in question-answering systems,this paper adopts deep learning-based natural language processing technology and combines it with medical domain knowledge to conduct related research on automatic question-answering based on Frequently-asked Questions(FAQ)standard library.Specifically,the following work is done:(1)Based on the FAQ question-answering system,this paper proposes a Chinese medical question matching method that combines Siamese recurrent neural networks and dual attention mechanisms to more efficiently capture semantic information,clarify the user’s questioning intention,and address the current problem of inaccurate acquisition of rich contextual semantic information in question text by existing models.To obtain more specific and complex text semantic information,this paper first uses pre-trained models to obtain two levels of embedding vectors,and then uses a fused neural network model to solve the question matching problem in the FQA medical question-answering system,effectively eliminating the impact of spelling errors,unrecognized synonyms,and other factors on matching results,and fully considering the importance of preceding and following sentences,optimizing weight allocation,and eliminating noise to more efficiently and accurately capture question semantic information.Experimental results show that the proposed model performs better than other mainstream neural network models in medical text matching tasks,effectively improving the effectiveness of Chinese medical text matching.(2)For non-FAQ question answering systems,answers matching the user’s questions need to be retrieved from massive medical information.A medical question answering matching method is proposed that introduces external knowledge.First,a pre-trained model is used to learn the features of the text and input them into a fused bidirectional GRU and attention mechanism module to better learn the information of key words in the question-answer pairs.Then,an attempt is made to add external knowledge from the knowledge graph to the data,and entity linking is performed between the dataset and the medical knowledge graph.Experimental results show that the introduction of attention mechanisms and external knowledge can effectively improve the performance of the model,and further reduce the interference of knowledge noise on the learning of key words.The experimental results show that the proposed method is effective in solving Chinese medical question answering matching tasks,achieving the highest experimental score compared to the baseline model.