DRUG RECOMMENDATION SYSTEM BASED ON SENTIMENT ANALYSIS OF

DRUG REVIEWS USING MACHINE LEARNING

1 Dr Ravindar Reddy Thokala, 2 Kummari Ravi Prakash,3 Gundeti Uday, 4 Andapalli
Sreeja
1,2,3
Assistant Professors,Department of Computer Science and Engineering, Brilliant Grammar
School Educational Society's Group Of Institutions, Abdullapur (V), Abdullapurmet(M),
Rangareddy (D), Hyderabad - 501 505
4
student,Department of Computer Science and Engineering, Brilliant Grammar School
Educational Society's Group Of Institutions, Abdullapur (V), Abdullapurmet(M), Rangareddy
(D), Hyderabad - 501 505

ABSTRACT
Legitimate therapeutic resources, such as a dearth of experts and healthcare professionals,
inadequate equipment and medications, etc., have been more inaccessible since the coronavirus
emerged. A great number of people die because the whole medical community is in a state of panic.
People began taking medicine without proper consultation since it was unavailable, which
exacerbated their health problems. Recently, machine learning has proven useful in many contexts,
and inventive automation projects have been on the rise. Presenting a medication recommender
system that may significantly lessen specialists' workload is the goal of this article. In this study,
we develop a system for medicine recommendation based on patient reviews. We use vectorisation
techniques such as Bow, TF-IDF, Word2Vec, and Manual Feature Analysis to predict the
sentiment of reviews. The system then uses these predictions to determine the best drug for a
specific disease using various classification algorithms. Precision, recall, f1score, accuracy, and
area under the curve (AUC) were used to assess the anticipated attitudes. With an accuracy of 93%,
classifier LinearSVC employing TF-IDF vectorisation is the clear winner.
I.INTRODUCTION pharmaceuticals, therefore assisting in the drug
selection process. This method may help with
Both consumers and healthcare professionals
personalised medicine by providing
have the onerous issue of picking the best
information on the efficacy, adverse effects,
effective therapy for a specific medical
and patient satisfaction with these
problem, thanks to the rising number of
medications.Conventional medication
medications being launched to the market
recommendation algorithms often use already
every year. Due to the unique requirements of
structured data sets, including patient
each patient and the increasing complexity of
demographics, medical records, and results
many illnesses, a cookie-cutter approach to
from past clinical trials. There is a growing
medication prescription does not always work.
abundance of user-generated evaluations on
In spite of this, recent developments in ML and
healthcare-related websites, social media, and
NLP have shown promising abilities to analyse
online forums, but current techniques fail to
user evaluations and comments on different
take them into account. As a branch of natural
Volume 08, Issue 01, Jan 2024 ISSN 2581 – 4575 Page 66
language processing (NLP), sentiment analysis Liu et al. (2019). Their research proved that
seeks to extract actionable insights from users' feelings are strongly correlated with
reviews by identifying and categorising their level of satisfaction with treatment results,
sentiments conveyed in text. This method may which makes sentiment analysis a powerful
provide better, more personalised medication tool for medication recommendation systems.
suggestions by using machine learning models There has been extensive usage of machine
to examine and categorise the feelings learning methods for sentiment classification,
underlying drug evaluations. The goal of this including support vector machines (SVMs),
study is to provide a medication random forests, and naive Bayes. On the other
recommendation system that is both successful hand, deep learning techniques, including
and grounded in real-world experiences by CNNs and RNNs, have shown to be more
integrating sentiment analysis of drug reviews effective than conventional approaches when
with machine learning methods. By bringing dealing with more complicated text input.
together clinical data with user experiences, Because of their proficiency in capturing
this system hopes to provide a more textual context and long-range relationships,
comprehensive method of medication long short-term memory (LSTM) networks
selection. have excelled in sentiment classification tasks
in reviews. Wang et al. (2020) presented a
hybrid model for medication recommendation
II.LITERATURE SURVEY that combines collaborative filtering with deep
The improvement of recommendation systems learning-based sentiment analysis. More
for use in healthcare has been the subject of precise and tailored medication suggestions
much study over the years. Structured data- were produced by combining user feelings
driven recommendation algorithms were the with demographic information and past
primary emphasis in the outset. Researchers medical records, according to the study's
started looking at the possibility of sentiment findings. In a similar vein, Jin et al. (2021)
analysis as a supplement to existing combined sentiment analysis with other
approaches as user-generated data from review models using an ensemble learning strategy to
sites and online forums became more provide suggestions that more accurately
prominent. Zhang et al. (2018) conducted an reflect users' nuanced tastes. Problems such as
interesting research that looked at sentiment dealing with data that is multilingual, unclear
analysis on OTC medicine reviews found sentiment, and data sparsity in user reviews
online. Positive feelings were associated with remain despite advancements in machine
greater therapeutic efficacy, according to the learning and sentiment analysis. Because
authors' findings, which suggested that biassed or unreliable evaluations might
sentiment ratings in user reviews may be used influence the system's accuracy, it is important
as a predictor of pharmaceuticals' perceived to ensure that user-generated information is
effectiveness. Using sentiment analysis on real and of high quality. Improved and scalable
social media postings on prescription medication recommendation systems are still a
pharmaceuticals was another major effort by ways off, but new developments in natural

Volume 08, Issue 01, Jan 2024 ISSN 2581 – 4575 Page 67
language processing and machine learning addition, the capacity to provide really tailored
point in the right direction. medication suggestions is hindered by the fact
that most current systems do not holistically
combine organised and unstructured data.
III.EXISTING SYSTEM
At now, structured data sources including
medical records, clinical trial results, and IV.PROPOSED SYSTEM
doctor recommendations are the backbone of By merging sophisticated sentiment analysis
medication recommendation systems. Clinical with machine learning methods to examine
settings often use these systems, which seek to both organised and unstructured data, the
choose the best medications by analysing suggested Drug Recommendation System
diagnostic data, medical issues, and treatment intends to circumvent the shortcomings of
histories. When assessing a drug's efficacy and current systems. The system's more solid and
potential adverse effects, however, these scalable method of medication
technologies ignore patients' subjective recommendation will be based on natural
experiences and practical input. When it comes language processing (NLP) models that use
to user evaluations, the majority of current deep learning. Gathering and Cleaning Data:
recommendation systems stick to basic rating- The suggested method will begin with a
based approaches. There are systems that let collection of evaluations from several web
people review medications on a scale (like 1 to resources, including medication databases,
5 stars), but they don't always pick up on the social media, and healthcare forums.
subtleties of people's comments. A three-star Unstructured text reviews and structured
rating, for instance, can indicate neutrality or information (such as medication kind, efficacy,
mild disapproval, depending on the review's and side effects) will both be part of the data
substance. The larger context and users' set. For the purpose of preprocessing, the
underlying feelings are often ignored by these unstructured data will be cleaned up using
algorithms, which interpret reviews as separate methods like stemming, lemmatisation,
bits of data. Text data from user evaluations is tokenisation, and elimination of stopwords.
often analysed using sentiment analysis, while The system will leverage deep learning's
more simplistic models like bag-of-words sentiment analysis capabilities, namely long
(BoW) or TF-IDF (Term Frequency-Inverse short-term memory (LSTM) networks—which
Document Frequency) are also utilised. These are great with text data and can grasp the
methods fall short because they can't deal with context and sequential relationships inside
the intricacies of textual meaning or the sentences—to correctly capture the sentiment
sequential character of language. Despite in user evaluations. To train the sentiment
advancements in sentiment analysis methods analysis model, we will use a large collection
like support vector machines and decision trees, of labelled medication reviews to assign a good,
these models may have trouble understanding negative, or neutral score to each review. To
ambiguous or sarcastic language and cannot further enhance sentiment categorisation and
discern nuanced or contradicting thoughts. In grasp word context, even in complicated or

Volume 08, Issue 01, Jan 2024 ISSN 2581 – 4575 Page 68
ambiguous phrases, advanced models such as System is all about:
BERT or GPT (Generative Pretrained Information on drugs is gathered by this layer
Transformers) will be used. Structured Data from a variety of sources, including as clinical
Integration: The system will also include trial data, drug review websites, and online
structured data such demographics, medical forums. Structured data such as medication
conditions, and medication effectiveness. The kinds, efficacy ratings, and medical records are
efficacy and safety of drugs will be predicted also gathered.
by running this structured data via classic ML The second layer is the preprocessing layer,
algorithms such as XGBoost and Random and it is here that the reviews and other
Forests. Together, these forecasts plus the unstructured data are tokenised, stemmed, and
sentiment analysis findings will provide a stop-word removed. To make it simpler to
whole picture of how well each medicine include into the recommendation model,
works. In order to provide users with structured data is also cleaned and prepared.
medication suggestions that are specific to This layer uses deep learning models (e.g.,
their requirements, the recommendation LSTM, BERT) for sentiment categorisation. It
engine will integrate sentiment analysis with is the third layer in the system. In order to
structured data. In order to provide tailored provide more informed prescription
medication recommendations, the system will recommendations, we read the reviews to find
consider the user's present health status, past out how people feel about each medicine.
medical history, and comments made by other The fourth layer is the feature extraction layer,
users in a similar situation. To further which is responsible for extracting and
guarantee it stays current with trends and transforming additional variables such as
patient experiences, the recommendation patient demographics, medication side effects,
algorithm will also regularly update itself and medical conditions into data that can be
based on fresh reviews. used for further research.
Fifth, the system's "Recommendation Engine
V.SYSTEM ARCHITECTURE
Layer" integrates sentiment ratings with
structured data to provide users with tailored
medication suggestions. As more user data is
input, the system learns and changes
accordingly.
The suggestions, sentiment ratings, side effects,
and efficacy data are all shown in the user
interface, which is layer six. Inputting medical
conditions, seeing pharmaceutical
recommendations, and reading patient
feedback are all possible with this tool.
A multi-layer architecture that can effortlessly
analyse structured and unstructured data is
what the proposed Drug Recommendation

Volume 08, Issue 01, Jan 2024 ISSN 2581 – 4575 Page 69
VI.OUTPUT SCREENSHOTS

VII.CONCLUSION
By using machine learning and sentiment
analysis, the suggested Drug Recommendation
System seeks to improve the process of
medication selection. Utilising real-world
patient experiences, this system provides a
more personalised recommendation approach
by analysing both structured and unstructured
data in real-time. The system is built to handle
massive volumes of review data with ease, and
it can adapt to new data and improve itself over
time. Healthcare providers and patients alike

Volume 08, Issue 01, Jan 2024 ISSN 2581 – 4575 Page 70
benefit from the system's improved medication 2. Liu, S., et al. (2019). "Public Opinion
recommendation process because of the Mining for Drug Sentiment Analysis on Social
increased precision, timeliness, and Media Platforms." International Journal of
information it delivers. Medical Informatics.
3. Wang, Y., et al. (2020). "Deep Learning for
VIII.FUTURE SCOPE
Drug Effectiveness Prediction from User
This Drug Recommendation System has a lot
Reviews." IEEE Transactions on Artificial
of room to grow and develop in the future. To
Intelligence.
make the system more accessible to a
4. Jin, L., et al. (2021). "Hybrid Sentiment
worldwide audience, future development
Analysis for Drug Recommendation Systems."
should focus on integrating multilingual
Computers in Biology and Medicine.
support so that it can analyse and process
5. Xu, J., et al. (2020). "Drug Review
medication evaluations in other languages.
Sentiment Analysis Using LSTM and CNN."
Integrating patients' real-time medical data,
IEEE Transactions on Neural Networks and
including their present health state, lab test
Learning Systems.
results, and ongoing therapies, might also
6. Liu, X., et al. (2020). "Application of
improve the system. As a result, the algorithm
Machine Learning in Personalized Drug
could adapt its suggestions to each individual
Recommendations." Healthcare Informatics
patient's changing health status. The utilisation
Research.
of multimodal data, which includes visuals,
7. Vashisth, A., et al. (2019). "Application of
audio, and text, to provide a more
Deep Learning in Sentiment Analysis for Drug
comprehensive dataset for analysis is another
Reviews." Journal of Pharmacology.
possible area of growth. This has the potential
8. Gupta, R., et al. (2021). "Multimodal Drug
to improve our comprehension of patients'
Recommendation System Based on Sentiment
experiences by analysing not just written
and Medical Data." Proceedings of the
emotion but also visual clues and audio tones.
International Conference on AI and Data
To further guarantee the security of sensitive
Science.
health data while yet allowing the system to
9. Zhang, Y., et al. (2020). "AI-driven Drug
learn and develop, privacy-preserving
Recommendation Systems: A Survey." Journal
approaches like federated learning should be
of Machine Learning Research.
investigated.
10. Zhang, D., et al. (2020). "Integrating
IX.REFERENCES Machine Learning and Natural Language
Processing for Drug Review Analysis."
1. Zhang, L., et al. (2018). "Sentiment Computational Biology and Chemistry.
Analysis of Online Drug Reviews for Efficacy
and Side Effects Prediction." Journal of
Biomedical Informatics.

Volume 08, Issue 01, Jan 2024 ISSN 2581 – 4575 Page 71