Research And Application On Detecting Adverse Drug Reactions From Social Media

At present,Adverse Drug Reactions(ADR)have become a hot issue for the medical community and the public.The issue of medication safety has increasingly attracted the attention of the whole society.It has been reported that diseases caused by adverse drug reactions accounted for 5% of the total inpatients,and mortality caused by adverse drug reactions ranked fifth in hospital mortality.The comprehensive knowledge of adverse drug reactions can provide a comprehensive understanding of adverse drug reactions and reduce their adverse impact on patients and the health system.Therefore,how to judge and predict adverse drug reactions has great theoretical value and practical value.In the past,drug safety supervision mainly relied on spontaneous reporting systems.These systems consist of suspected ADR reports collected from healthcare professionals,pharmaceutical companies,and are primarily maintained by regulatory agencies and health agencies.The American Food and Drug Administration's(FDA)Adverse Event Reporting System(AERS)is one such system.However,recent studies have shown that AERS seriously underestimates the incidence of ADR.This spontaneous reporting system has also been subject to a series of restrictions,including inadequate reporting(approximately 10% of ADR with severe ADR),excessive known reports,incomplete data,duplication of reports,unclear causality,and so on.Therefore,the focus of recent research has expanded to use other data sources to detect ADR.Social media platforms(such as Twitter)are emerging digital communication channels that provide an easy way for ordinary people to share their health and medication experiences online.As more people openly discuss their health information online,social media platforms provide a rich source of information for exploring adverse drug reactions.Therefore,the main task of this dissertation is to effectively identify the texts related to adverse reactions in Twitter,determine whether a certain tweet contains references to adverse reactions,and identify entities that mention adverse reactions.In the mentioned task of adverse drug reactions,we construct a convolutional neural network model of a converged network structure,transforms the text into a representation based on a distributed vector,and effectively identifies the text referring to adverse drug reactions in Twitter.In the task of identifying adverse reaction entities,a large number of unlabeled data are used to learn the low-dimensional,dense word vectors,and then BiLSTM is used to process the word vectors and word character vectors of documents,extracting the contextual representation of each word,and then the full-text range The internal context representation and the context's proximity of the word are sent to the CRF layer after fusion,and finally the CRF layer is used to obtain the tag sequence corresponding to the entity.This article provides valuable reference information for the further study of adverse drug reactions and reduces the consumption of money and time in the discovery of adverse drug reactions.