With the development of biological medicine,a large number of biomedical literatures have been produced.Massive literature contains a wealth of cutting-edge biomedical knowledge.Also it is an important source of knowledge medical personnel.Therefore,it is important to obtain the knowledge we need from the machine learning.Drug-drug interaction extraction refers to the interaction between two or more drugs in the extraction of medical text.Drug warning is the issue when there is an interaction between drugs.For chronic disease patients,who take drugs for a long time,the drug warning database protects them healthy.This thesis studies and analyzes the model of drug-drug interaction extraction and the algorithm of text relation extraction in the current stage.According to the massive information of medical professional vocabulary,high importance of text information and the unideal feature selected in this stage,this thesis proposed a dynamic sorted convolutional neural network(DSCNN)to improve the performance of drug-drug interaction extraction.At the same time,the pre-training word vector is used to extract the initial information of the text to improve the reliability and scalability of the algorithm.Finally,the performance of the text is improved and the feasibility of the early warning program for chronic diseases.Both of them can be verified in the thesis.In the 2013 DDIExtraction task data sets,the dynamic ordered convolutional neural network algorithm is used to improve the pressesion rate.In the aspect of drug warning of chronic diseases,this thesis designs a method based on machine learning,which uses the word vector transformation of skip-gram model and the visualization of T-SNE dimension reduction.A drug warning system for chronic diseases was established by tensorflow,and the results were visualized with tensorboard.The results of this study have positive significance for drug-drug interaction extraction,drug warning and auxiliary diagnosis. |