System for Recommending Drugs Based on Machine Learning Sentiment Analysis of Drug Reviews

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 07 | July 2022 www.irjet.net p ISSN: 2395 0072

System for Recommending Drugs Based on Machine Learning Sentiment Analysis of Drug Reviews

1Student, Dept. of CSE, Rajeev Institute of Technology, Hassan, Karnataka, India

2Professor and Head, Dept. of CSE, Rajeev Institute of Technology, Hassan, Karnataka, India ***

Abstract - Since the coronavirus emerged, it has been increasingly difficult to get legitimate therapeutic resources, such as the scarcity of specialists and healthcare professionals, appropriate equipment and medications, etc. There are many deaths as a result of the medical profession as a whole being in turmoil. Due to a lack of availability, people began taking medication on their own without the proper consultation, which made their health situation worse than usual. Recently, machine learning has proven useful in a variety of applications, and creative work for automation is on the rise. The purpose of this paper is to present a system for prescribing drugs that can considerably minimize the workload of experts. Thus, a proposed system provides a drug recommendation platform that requires patient feedback to predict sentiment using a number of vectorization techniques.

Key Words: Drug Recommendation, Machine Learning, SentimentAnalysis,Reviews,Ratings.

1. INTRODUCTION

Asaresultofthemassiveriseincoronaviruscases,thereisa worldwidedoctorshortageparticularlyinruralareaswhere therearefewerexpertsthaninurbanareas.Adoctormust complete their education between six and twelve years. Therefore,itisimpossibletoincreasethenumberofdoctors inashortamountoftime.AninfrastructureforTelemedicine needs to be pushed up as soon as possible in this difficult moment. Medical errors occur often today. Over 200 thousandpeopleperyearinChinaand100thousandpeople peryearintheUSAareassociatedwithdrugerrors.More than40%ofthetimespecialistsmakeerrorswhilewriting prescriptions,becausespecialistsonlyhavealimitedamount ofknowledgetobasetheirdecisionson.Choosingthebest medication is essential for people who need specialists throughin depthunderstandingofmicroscopicorganisms, antiviral medications, and consumers. Every day, new research is produced, and more medicines and testing equipment are made available to medical professionals. However,selectingacourseoftreatmentorprescriptionfor a particular patient based on reasons and early clinical historyprovestobeevermoredifficultforclinicians.Item reviewshaveevolvedintoacrucialandessentialaspectfor purchasing things globally due to the internet’s rapid expansionandthegrowthoftheweb basedcompanysector. Before making a purchase, people all over the world have grown conventional to reading comments and searching

internet.Althoughthemajorityofpriorresearchfocusedon evaluatingexpectationsandproposalsfortheareaofhealth care or scientific therapies, E Commerce sector has been infrequently examined. The amount of people searching online for a diagnosis because they are concerned about theirhealthhasincreased.APewAmericanResearchCenter surveytakenin2013foundthataround35%ofindividuals checkedfordigitalclinicaldiagnosis,whereasabout60%of peoplesearchedforhealth relatedissues.Inordertohelp physicians and patients learn more about medications for specific medical situations, a medication recommender system is obviously essential. A common method called a “recommenderplatform”suggestsitemstousersbasedon theirneedsandbenefits.Theseapproachesusecustomer’s reviewstoanalyzeclientemotionandofferasuggestionfor theirspecificrequirements.Thedrugrecommendersystem uses sentiment classification and feature extraction to conditionally provide medications depending on patient feedback. Sentiment analysis is a development of approaches,techniques,andinstrumentsforidentifyingand separatingsentimentalinformationfromlanguage,suchas opinions and thoughts. On the other hand, Featuring engineeringrequiresaddingnewfeaturestotheonesthat alreadyexistinordertoenhancemodelperformance

2. PROBLEM STATEMENT

The world is experiencing a doctor shortage due to the exponential increase in coronavirus cases, particularly in ruralareaswheretherearefewerspecialiststhaninurban areas.Adoctormustcompletetheireducationbetweensix andtwelveyears.Asa result,itisimpossibletoaddmore doctors in a short amount of time. An infrastructure for Telemedicineneedstobepushedupassoonaspossiblein thisdifficultmoment.

3. LITERATURE SURVEY

WittichCMetal.[1]Theworkinthispaperisfocusingonthe pharmaceutical errors which are reviewed for practicing physicians with an emphasis on terminology, definitions, incidence, risk factors, disclosure and legal consequences. Numerous variables can contribute to medication errors, including those related to the drug, the patient, and the healthcare provider. One or more of the outcomes that doctors may encounter after making drug prescription errors includes losing the faith of their patients, civil

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 07 | July 2022 www.irjet.net p ISSN: 2395 0072

proceedings,criminalcharges,andhealthboarddiscipline. Variousapproacheshavebeentriedwithvaryinglevelsof successinpreventingpharmaceuticalerrors.Theabilityof medicalprofessionalstogivetheirpatientssafecaremaybe improvedbylearningmoreaboutdrugerrors.

Bartlett JG et al. [2] In more than 10 years since the last Community AcquiredPneumonia(CAP)proposalfromthe American Thoracic Society (ATS) / Infectious Diseases SocietyofAmerica,theprocessforcreatingguidelineshas altered,andnewclinicaldatahavebeencreated(IDSA).Due to the expansion of information regarding the diagnostic, treatment, and managerial decisions for the patient care with CAP, we purposefully limited the extent of this framework to cover judgments from the point of medical diagnosis ofpneumonia tothe end ofantibiotic treatment andcarrychestimageprocessing.

T.N.Tekadeetal.[3]Thisarticleoffersabriefsummaryof aspectminingmethodsastheyapplytothesearchfornew drugs. For the pharmaceutical industry, it is crucial to conduct research on the earliest possible detection of adverse drug reactions. A difficult task is identifying important topics from brief and noisy reviews. The probabilisticaspectminingmodel(PAMM)issuggestedasa solutiontothisissueinordertofindtheaspectsandsubjects related to class labels. Due to a special characteristic of PAMM,itconcentratesondiscoveringfeaturesspecifictoa singleclassratherthansimultaneouslydetectingfeaturesfor allcategoriesduringeachoperation.

Doulaverakis et al. [4] Drug drug and drug disease interactions can be difficult to identify, and finding the necessary information can be challenging due to the enormous number of medications that are already on the marketandtheongoingpharmaceuticalresearch.Although international standards have been created to facilitate effective information exchange, such as the ICD 10 classificationandtheUNIIregistration,medicalstaffstillhas toberegularlyinformedinordertoefficientlyidentifydrug interactionspriortoprescription.Inpriorpublications,the usageofSemanticWebtechnologyhasbeensuggestedasa solutiontothisissue.

Gao,Xiaoyanetal.[5]Theworkinthispaperisfocusingon the recommendation of drugs with Graph Convolution Network, which mainly employs the mechanism of informationpropagationandembeddingpropagationlayers to model high order connectivity and elaborate the representationlearning.Theproposedsysteminvolvesthree keycomponentsnamelytheembeddinglayer,information propagation, and prediction layer. The work is mainly focuses on the accuracy rather than the evaluation of the recommendationsystem.

Li ChihWangetal.[6] Theproposedsysteminthispaper focusedon recommendinga parameterthatisefficient by

using a curing parameter recommendation system. The proposedsysteminvolvesavotingmethodthatisdeveloped bysevenMachinelearningalgorithms.TheseMLmodelsare trainedastheclassifiersmainlytorecommendacandidate representative medical datasets. The dataset with the highest frequency is chosen to be the recommended representativedataset.Longshort termmemorynetworks areusedforpre adjustingtopredicttheheatingcurve.

Susannahetal.[7]Inthisresearch,adeeplearningapproach for health based medical datasets is proposed. This approach automatically identifies what meal should be suppliedtowhichpersondependingontheconditionand other parameters like age, race, body weight, calories, fat, sodium, protein, fiber and cholesterol. The integration of deep learning and machine learning methods such as regressionanalysis,naivebayes,recurrentneuralnetworks, multilevel perceptrons, gated recurrent units, and long short term memory (LSTM) is the main goal of this study framework.ThecharacteristicsoftheseIoMTsampleswere evaluated and further processed prior to using machine learning,deeplearning,andotherlearning basedmethods.

4. OBJECTIVES

1.Analyzingtheinternalworkingsofourproposedsystem usingindividualandgroupmachinelearningtechniqueslike regressionanalysisandnaivebayes,aswellasdeeplearning algorithmslikeGRU,RNNandLSTM

2. Giving a thorough explanation of how our system functionsinaccordancewiththeproductandpatientdisease specifications.

3.ExamininghowourAIanddeeplearningsystemsbehave inordertobettergraspthenatureofthepatient’sproblems andthemedicationstheyshouldtakeattherighttime.

4.Wedemonstratedusingastudyofourmachinelearning and deep learning that various patient conditions have various recommender proofs, which may call for various treatmentsandparticularcare.

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 07 | July 2022 www.irjet.net p ISSN: 2395 0072

5. SYSTEM ARCHITECTURE

C. Model Construction and Model Training

TheactoftraininganMLmodelinvolvesprovidingthe learning algorithm with training set to use as a learning resource. The model artifact produced during training is recognized as a “Machine Learning model”. The correct solutionsometimesreferredtoasagoalortargetattribute, needstobeincorporatedinthetrainingdata.Thelearning method constructs an ML model that represents these patterns by looking for patterns in the training data that relatethecharacteristicsoftheinputdatatothetarget.

D. Model Verification and Outcome Evaluation

Figure 1: RepresentationofSystemArchitecture

Theabovefiguredeterminesthesystemarchitectureofthe proposed system. The system architecture involves followingsteps:

A. Data Gathering and Preprocessing

Machine Learning needs models and a lot of data in order to work. The procedure of gathering signals that monitor actual physical situations and converting the obtained results into electronic integer values which a computer can manipulate is known as Data Gathering. Processing primary data involves the subsequent procedures.Itisnecessarytocombineahugeamountofraw data obtained from field surveys in order to compare the detailsofindividualresponses.Amethodfortransforming unclean data into clean data sets is known as Data Preprocessing. Real world information is consistently inaccurateandmissingofparticularbehaviorsorpatterns.It isalsofrequentlyinconsistentandincomplete.

B. Feature Selection and Data Preparation

In order to create attributes for machine learning algorithms,onemustusedomaininformationfromthedata. The technique used here is called feature engineering. By generating features from input data that assists in the machinelearningmodel,featureextractioncanimprovethe prediction capacity of machine learning algorithms. In machinelearning,featureengineeringistheessentialskill that distinguishes significantly among a successful model and a poor model. The concept of “feature engineering” involvestakingrawdataandturningitintofeaturesthatthe predictive models can use to more accurately depict the underlyingissue.Thepracticeofgroupingandcategorizing data based on particular characteristics is known as Data Classification. It may be done either in accordance with numericalcharacteristicsorinaccordancewithattributes.

Themodelisemployedtofreshinputduringthetesting phase.Therearetwodistinctsamplesforthetrainingand testdata.Designingamachinelearningtechniquewiththe intentionofperformingiteffectively.Generalizewelltofresh datainthetestsetaswellasthetrainingset.Real timedata willbepassedforthepredictionwhenthebuiltmodelhas been evaluated.Oncea forecasthasbeenmade,theresult willbeexaminedforthemostimportantdata.

6. RESULTS

Thedrugreviewsampleutilizedinthisstudywasobtained from the UCI ML resource. This data comprises six components:thenameofthedrugtaken,thereviewofthe patient, the patient’s status, the valuable count, which indicatestheamountofpeoplewhoencounteredthereview beneficial,thedateofthereviewentry,anda10 starpatient ratingthatindicateshowsatisfactorythepatientisoverall. Accordingtotheuser’sstarrating,eachreviewinthiswork was categorized as either positive or negative. Positive ratingsarethosewithfiveormorestars,whereasnegative ratingsvaryfromonetofivestars.

In figure 2 we can see top 20 medical conditions with the greatestnumberoftreatmentoptions.Onefactortoobserve inthisfigureisthattherearetwogreenbars,whichindicate thecriteriathathavelittlesignificance.

Figure 2: Mostrecommendeddrugsperconditions

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Figure3demonstratesthat thefeaturesusedareeffective patternpredictorswithhighaccuracyandlittleerror.

93 percent accuracy. On the other hand, the Word2Vec Decisiontreealgorithmscoredtheworst,reachingonly78% accuracy.Weintegratedthetopexpectedsentimentvalues fromeachstrategyLGBMonWord2Vec(91%)Perceptron on Bow (91%) Random Forest on manual features (88%) LinearSVC on TF IDF (93%) and combined them by the standardized useful count to establish a recommender system.Thisprovideduswiththedrug’stotalscoreforeach condition. In order to enhance the effectiveness of the recommender system, future work will evaluate various oversamplingtechniques,usealternativen gramvalues,and simplifyalgorithms.

REFERENCES

Figure 3: Modelaccuracyvsloss

Figure4displaysthetopfourmedicationsthatouralgorithm recommendsforthefivetopmedicalissuesincludingacne, contraception,highbloodpressure,anxietyanddepression.

[1] WittichCM,BurkleCM,LanierWL.Medicationerrors: an overview for clinicians. Mayo Clin Proc. 2014 Aug;89(8):1116 25.

[2] BartlettJG,DowellSF,MandellLA,FileTMJr,Musher DM,FineMJ.Practiceguidelinesforthemanagementof community acquiredpneumoniainadults.Infectious Diseases Society of America. Clin Infect Dis. 2000 Aug;31(2):347 82.

[3] T.N.Tekade and M.Emmanuel,“Probabilistic aspect miningapproachforinterpretationandevaluationof drugreviews,”2016InternationalConferenceonSignal Processing, Communication, Power and Embedded System(SCOPES),Paralakhemundi.

[4] Doulaverakis, C., Nikolaidis, G., Kleontas, A. et al. GalenOWL: Ontology based drug recommendations discovery.JBiomedSemant3,14(2012).

[5] Gao,Xiaoyan,FuliFeng,HeyanHuang,Xian LingMao, TianLan,andZewenChi.“Foodrecommendationwith graph convolutional network.” Information Sciences 584(2022):170 183.

Figure 4: Topfourmedicationsproposedforthefive majorconditions

7. CONCLUSION

Whether we are purchasing, buying products online or eating out, reviews are gradually becoming a part of our daily routine. We use reviews to help us make the best choices.MultipleMachineLearningtechniqueswereusedto construct a recommender system which includes Perceptron,Multinomial NaiveBayes,Logistic Regression, RidgeclassifierandLinearSVCimplementedonTF IDF,Bow andclassifierslikeLGBM,DecisionTree,andRandomForest. Ourexaminationofmodelsusingfivemainmetrics:f1score, validity, recall, precision and AUC score reveals that the LinearSVCusingTF IDFoutperformsallothermodelswith

[6] Chen, Yu Xiu, Li Chih Wang, and Pei Chun Chu. “A medicaldatasetparameterrecommendationsystemfor anautoclaveprocessandanempiricalstudy.”Procedia Manufacturing51(2020):1046 1053.

[7] Fox, Susannah and Duggan, Maeve. (2012). ImplementingaMachineLearningModeltoRealizean EffectiveIOMT AssistedClientNutritionRecommender System.PewResearchInternetProjectReport.

[8] J. Ramos et al., “Using tf idf to determine word relevanceindocumentqueries,”inProceedingsofthe firstinstructionalconferenceonmachinelearning,vol. 242,pp.133 142,Piscataway,NJ,2013.

[9] N. V. Chawla, K. W. Bowyer, L. O. Hall and W. P. Kegelmeyer.SMOTE:SyntheticMinorityOver sampling Technique, 2011, Journal of Artificial Intelligence Research,Volume16,2020.

Volume: 09 Issue: 07 | July 2022 www.irjet.net p ISSN: 2395 0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal |

International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056

Volume: 09 Issue: 07 | July 2022 www.irjet.net p ISSN: 2395 0072

[10] LeileiSun,ChuanrenLiu,ChonghuiGuo,HuiXiong,and YanmingXie.2016.Data drivenAutomaticTreatment Regimen Development and Recommendation. In Proceedings of the 22nd ACM SIGKDD International ConferenceonKnowledgeDiscoveryandDataMining (KDD’16).AssociationforComputingMachinery,New York,NY,USA,1865 1874.

[11]V.Goel,A.K.GuptaandN.Kumar,“SentimentAnalysis of Multilingual Twitter Data using Natural Language Processing,” 2018 8th International Conference on Communication Systems and Network Technologies (CSNT),Bhopal,India,2018.

[12] Shimada K, Takada H, Mitsuyama S, et al. Drug recommendation system for patients with infectious diseases.AMIAAnnuSympProc.2005;2005:1112.

2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal