Design And Implementation Of Medical Retrieval System Based On Deep Learning

In the big data environment,the amount of Internet data is increasing,and the retrieval system can help people quickly find the information that meets their needs from the massive data.At the same time,these massive data promote the rapid development of deep learning related technologies.In recent years,pre-training technology has developed rapidly and has been widely used in various deep learning tasks.Most traditional text retrieval systems use vector space model,which does not consider word order and context information.The pretraining model trained by a large number of corpora can abstract the semantic information of the text and apply it to the retrieval system to effectively improve the accuracy of the retrieval.Due to the particularity of the field,medical related information needs high accuracy.Especially since the outbreak of COVID-19,people are in great need of common medical related information,but there is a lack of effective ways.Therefore,it is of great significance to develop a system that can quickly and accurately retrieve common medical related information.This thesis combines indexing technology,deep learning model,pre-training and other technologies to design and implement a medical retrieval system based on deep learning,through which users can input common medical related queries to retrieve relevant documents in the corpus.The core of this thesis is to apply LSTM network.siamese network and BERT pre-training technique to retrieval system,and realize siamese BiLSTM network model by reforming these networks.This thesis also compares the effects of traditional Word2Vec,Fasttext and pre-training techniques on vectorization of dataset.Redis cache technology is used in this thesis to improve the retrieval efficiency.The experimental results show that the medical retrieval system can achieve a very good retrieval effect on the test set and meet People’s Daily needs.The medical retrieval system consists of four modules:1.Data processing module:including timing data crawling,data integration and data preprocessing.2.Index processing module:including building indexes on integrated data sets and incremental indexes on data that are periodically climbed.3.Web service module:This thesis adopts Flask to build the background of the system,and designs the front end of the system by combining wechat public account,HTML,JavaScript,AJAX and other technologies.4.Retrieval module:including offline training and evaluation of the model;Query correction,correct the wrong query keywords;Cache processing,used to reduce the response time of the system to process similar queries;Index query,through the index query document set and user query related documents;Model sorting:Use the trained model to sort related documents according to their similarity to user queries.The retrieval speed of this medical retrieval system can reach second level.It has certain practicability and expansibility in function.In the realization of the use of the current mainstream technology,can provide users with good retrieval services.