Study On The Model Of Knowledge Discovery And Utilization In Adverse Drug Reaction

Adverse drug reactions are dangerous to human health.Restricted to the clinical trials,the number of clinical trials,the duration and the characteristics of the trials,it is difficult to obtain complete information about the drug safety before a drug's launch.So the post-marketing surveillance becomes particularly important.The social media provides a new way to collect and make use of the medical data.This paper aims to improve the scientific and intelligent level of pharmacovigilance with the help of social media,and to provide referential methods for the post-marketing surveillance and personalized health information service.For that reason,the advanced algorithm of Natural Language Processing,machine learning,data mining,artificial intelligence and semantic analysis is applied to the discovery and utilization of the adverse drug reactions based on social media,and thus achieve the social media big data filtering,signal extraction of adverse drug reaction,the adverse drug reaction ontology construction,and the early warning of adverse drug reaction.This paper is the study of theory and approach for the application.The full text includes nine chapters.Apart from the introduction and the summary,the following outlines the rest of the article:Chapter 1:The clarification of the related theory.First of all,it discriminates the concepts of the adverse drug reaction,which is the research of this paper.Secondly,it teases out the knowledge discovery and defines the knowledge discovery in adverse drug discovery.Thirdly,it reviews the network healthy community,which is the research object of this paper.Finally is the DIKW theory,which is the theory basis as well as threshold of the general layout of this paper.Chapter 2:The general clarification of the model of knowledge discovery and utilization in adverse drug reaction based on the DIKW theory.First,it comes up with the basic elements of the knowledge discovery and utilization of adverse drug reaction,including the adverse drug reaction big data,adverse drug reaction knowledge and adverse drug reaction intelligence,and then defines them in detail.Secondly,it reveals the information activities that are involved in the knowledge discovery and utilization of adverse drug reaction,including the filtration of the adverse drug reaction big data,the signal extraction of the adverse drug reaction,the semi-automatic adverse drug reaction ontology construction and the personalized precaution of adverse drug reaction.Finally,it puts forwards the research model and reveals the information activity routines in the "data-information-knowledge-intelligence" information chain of adverse drug reaction.Chapter 3:The filtration of adverse drug reaction big data.It puts forward a framework that is about the filtration of adverse drug reaction relevant posts from the social media big data by means of the classifications in our paper.The framework consists of three parts,the mechanism of dimension reduction,automatic expanded training data and a resulting classifier.The classifier can effectively extract posts related with adverse drug reaction posted by the drug user from the health social media;in the meanwhile it can remove the posts that are from medical research and medical news,or the posts that are heard or copied from the others rather than the drug user.The modeling learning process is a semi-supervised classification,which helps to reduce dimension with the LDA model,to retrieve relevant unlabelled posts,and to filter report resource with the help of word bag feature extraction and the transductive support vector machine.Chapter 4:The signal extraction of adverse drug reaction.It puts forwards an adverse drug reaction signal extraction framework that combines statistical learning and semantic filter.First it recognizes the medical entity from the noisy social media based on multi-dictionary sources matching,including the extraction of UMLS standard medical entity,the filtration of the FAERS vocabulary,the expansion of the CHV user healthy vocabulary and so on.Then it applies statistical learning based on the shortest dependency path kernel to the extraction of the adverse drug reaction.Finally in order to improve the accuracy of the signal extraction of adverse drug reaction,it filters the pharmacotherapy,the applicable disease information and the negative adverse drug reaction by means of the semantic knowledge in the drug safety database.Chapter 5:The semi-automatic construction of adverse drug reaction ontology.First,based on the design concept that separates the business level and language level,it shows adverse drug reaction in the network health community in the form of"object—attribute—description".The fine granularity is reflected in the diversity of describing the same adverse drug reaction.Secondly,it puts forward a framework of semi-construction of adverse drug reaction ontology,which is based on deep learning.It automatically extracts the concepts in the adverse drug reaction ontology based on the word2vec candidate words extraction algorithm.And the fine-grained adverse drug reaction ontology provides a database for mining signal of potential adverse drug reaction by making use of social media.Chapter 6:The personalized precaution of adverse drug reaction.It applies the population characteristic to the ontology above,and creates an adverse drug reaction ontology based on the user story.It comes up with the ontology rule learning,which is to mine the implicit rules under the dataset according to machine learning and the data mining algorithm,and thus improve the efficiency of ontology generation.It can carry out the association rules mining on the adverse drug reaction according to different drug users' population characteristics.Chapter 7:The experimental analysis.This chapter verifies each link of the model respectively.In the experiment of 7.1,we verify the framework suggested above with three drug use reviews from the WebMD.The results show that in most cases,the ADR filtering method is obviously better than the baseline method,especially when the available ADR data is sparse.In the experiment of 7.2,we identify the medical entity and extract the adverse drug reaction event relationship from the data obtained from Diabetes Forums in sentences.Comparing our method with the co-occurrence analysis,we find that the method in this paper can dramatically improve the precision rate and the F-measure.In the experiment of 7.3,taking the diabetes drug as an example,we extract the fine-grained adverse drug reaction candidate words based on word2vec,and realize the semi-automatic construction of fine-grained adverse drug reaction ontology.In the experiment of 7.4,we apply the user story attribute to the ontology that was built in the last chapter.According to the association rules analysis of the hypoglycemic drug,we mine the association between the drugs,the patients' attributes and the adverse drug reaction,which we regard as the rules of ontology inference to realize the personalized precaution of the adverse drug reaction.