Mention Classification And Drug Interaction Classification Based On Adverse Drug Reactions

In today’s society,a series of new prescription drugs are constantly being put into the medical market.If there is no necessary supervision,Adverse Drug Reactions(ADR)caused by these drugs may threaten public health safety.A study report in 2000 showed that about 7,000 patients died of drug ADR every year.However,at present,most of the after-sales monitoring of drugs rely on official databases,but the existing reporting system on ADRs has omissions.Studies have shown that about 94% of ADRs are misreported on the official system.Therefore,in order to solve the limitations of the official ADR reporting system,some scholars have proposed to use the data on Twitter for ADR extraction to enrich the data sources to solve this problem.Twitter publishes more than 2 million posts per hour,providing a variety of rich multi-modal data,and they themselves provide researchers with many research opportunities.This social media platform also covers a wealth of health-related data samples;the content of the post contains a variety of health-related information,including disease history,disease symptoms and ADR.Twitter provides important health-related data for scientific research,and researchers can use this information to understand how patients feel about medication.This paper proposes the use of graph convolution(GCN)network to classify social media texts to efficiently obtain ADR information from tweets on social platforms(Twitter).This model can correctly identify the "word" level and "phrase" level ADR in tweets,and solves a common problem in the existing sequential learning model-the model is difficult to extract remote relevant information from the text.Based on the existing Text GCN model,this paper integrates emotional information and medical information in the figure according to the characteristics of social media and medical texts.Without using external word embeddings,the model was compared with multiple models on two Twitter data sets,and good results were obtained.It also effectively reduces the need for pre-training word vectors and reduces the time used to collect pre-training samples.For the task of dealing with drug-drug interaction(DDI),this article uses BERT as the basic model for the study,combines entity information and integrates the Gaussian distribution model.On the original basis,the model can obtain entities and entities.The surrounding information vector.The model uses the [CLS] vector representing all text information and the word vector fusing the entity and its surrounding information to join together,and then input the fully connected layer for multi-classification.This research provides some valuable references for the continued research of ADR.By enabling the model to be classified independently,it reduces the excessive consumption of money and labor costs in the process of ADR discovery.