2023 5th International Conference on Sustainable Technologies for

Industry 5.0 (STI), 09-10 December, Dhaka

Advancing Semantic Web Technologies: Enhancing

Data Interoperability, Knowledge Representation,
and Intelligent Recommendation Systems

Abstract— This paper proposes a deep learning system for tients. Our study employs a compound data set, which is
sentiment analysis and recommendation in the context of the derived from several sources, such as the UCI Machine
Semantic Web. By incorporating user reviews, ratings, and Learning Repository and Kaggle. It is a collection of user
metadata from various sources, the proposed model can analyze reviews of drugs and related information, such as the
sentiments and recommendations and drug recommendations condition for which the drug was prescribed, rating, date of
suitable for individual customers. These include text cleaning the review, etc. This broad data set ensures that the patients’
and n-grams generation to ensure the data is clean and in the feelings about their conditions and the efficiency of the
correct format. The deep learning model obtained a high
various drugs can be well understood. We cleaned the data to
accuracy of 89.92%. The comparison with other machine
learning models like LightGBM, XGBoost, LSTM, and Random
ensure that it was in the correct format and of high quality.
Forest showed the effectiveness of the proposed method. This involved pre-processing of the review texts, removal of
Furthermore, the model’s performance for different datasets and stop words and converting the text into a format suitable for
the user-friendly interface also suggest the potential of the model analysis by the use of n-gram [5]. Further, we excluded
to revolutionize the Semantic Web. This research helps to fill the conditions that had only one recommended drug to make
gaps in the existing literature on sentiment analysis and sure that the recommendations are quite strong. This
recommendation and provides a reliable solution for real-time preprocessing step is important for the model to learn how to
applications. generalize over different datasets and corpora, which will be
very useful in the different fields of healthcare.
Keywords— Semantic Web, Deep learning, NLP, CNN,
Recommendation To classify the sentiments in drug reviews we employed
the feature n-gram and trained a neural network to identify
I. INTRODUCTION instances of varying sentiment [6]. Furthermore, the model
The use of online forums and review websites has becom was developed to give either positive or negative label
e quite common. It has led to a large amount of information c depending on the review rating which was effective;
reated by users regarding the usage of some medications and especially in analyzing sentiments. In order to make the
their efficacy in treating certain diseases. This abundance of recommendation of drugs the predictions were then
data offers a significant chance to use NLP and machine lear employed multiplying the sentiment and the review rating
ning to examine patient attitudes and provide meaningful sug together in average [7]. This method makes sure that
gestions [1]. This paper employs the mixed dataset to build a recommended drugs are effective in the management of the
sentiment analysis and recommendation model as a proof of diseases and are not terrible to the patients. The modularity
concept to advance the application of big data analytics in pe of the model also makes it a flexible model that can support
rsonalized medicine. The vision of the Semantic Web, as pro several datasets in the health care application as well as other
posed by Tim Berners-Lee, is to establish a global platform f forms of patient generated data fine-tuned for specific use in
or sharing knowledge by providing data with precise semanti the Semantic Web.s
cs and making computers and people work together. Moreov II. RELATED WORKS
er, it expands the current web by encouraging the web of data
characterized by the shared data formats and exchange protoc The area of the Semantic Web combining with machine learn
ols on the web, mostly essentially RDF (Resource Descriptio ing is still active in the research field, especially in the contex
n Framework). semantic web means that data from different s t of the healthcare systems. It has been found in the latest lite
ources can be connected, thus making the web more connecte rature that these technologies can be used for improving data
d and meaningful [2]. In the field of healthcare, the Semantic exchange, semanticization, and decision making across differ
Web will enable the amalgamation and harmonization of diff ent fields and sectors such as agriculture, healthcare, and ind
erent types of health data and can improve the decision-maki ustry. Lynda et al. (2023) have offered a detailed article on th
ng process and the development of unique treatment regimen e use of machine learning in agriculture, which underlined th
s [3]. For instance, the combination of EHRs, genomics, and e necessity of using machine learning with semantic web tech
data from the patient’s social media profiles would be benefi nologies to address the issues of data heterogeneity and sourc
cial for healthcare practitioners. By integrating the machine l e complexity [8]. Likewise, Suzuki & Elgort (2024)
earning models into the Semantic Web system, the capability described how machine learning can build on the Semantic
of analyzing and interpreting the patient data can be extended Web by improving the automaticity of semantic annotations
and enhanced to provide more precise and related suggestion and the reasoning capacity over extensive data [9]. Hafidi et
s [4]. This integration means it is not only possible to find the al. (2023) discuss the present and future of Semantic Web
information but also to use it, and it brings the prospect of int Technologies, with a focus on the fact that there is a need for
elligent systems that will help both healthcare workers and pa sound frameworks for integrating both structured and

979-8-3503-9431-3/23/$31.00 ©2023 IEEE

unstructured data. They also pointed out that Knowledge & 15]. These measures help to minimize the level of bias
Graphs (KGs) are critical in realizing this integration, which inherent in using sentiment analysis and achieve a higher
is in line with our use of hybrid datasets [10]. Kulmanov et degree of objectivity and fairness of the results obtained.
al. (2021) have expanded on the integration of semantic
similarity measures with machine learning and suggested III. WORKING METHOD
ways to improve bioinformatics solutions using ontologies Building on the synergistic use of NLP, deep learning, an
and semantic Web technologies [2]. In recent years, the use d Semantic Web technology, we created a very comprehensi
of deep learning for sentiment analysis has been receiving a ve system for sentiment analysis and drug prescription for thi
lot of attention, especially in handling user-generated content s study. This paper has outlined steps relating to data acquisit
across different domains. The recent work of Yahya et al. ion and preparation, model creation, and assessment, as highl
(2021) focused on the application of semantic web ighted in the following sections.
technologies and knowledge graphs in Industry 4. 0,
acknowledging that deep learning is capable of enhancing the
precision and detail of the data collected [11]. The semantic
web was described by Hitzler (2021) as a field that has
evolved and advanced with the need for real-time
applications, especially when it comes to data that can be
changed dynamically through deep learning [12].
However, to the best of our knowledge, the research on
automating drug recommendations through sentiment
analysis is still limited and far from maturity. Previous works .
are mostly based on theoretical models or back-end
computations without taking into account the dynamic user
activity. In the methods and models for semantic memory,
Kumar (2021) noted that more research into interactive A. Dataset & Modificationss
systems and methods for generating a real-time response for
Our study utilizes a hybrid dataset collected from multipl
users is lacking [13]. The literature review shows that the
e sources, including the UCI Machine Learning Repository a
application of machine learning in conjunction with semantic
nd Kaggle. This dataset comprises user reviews of various dr
web technologies, including KGs and NLP, in the healthcare
ugs, along with metadata such as drug names, conditions trea
domain has made considerable progress in the current state of
ted, ratings, review dates, and useful counts. To enhance the r
the art. Still, several gaps remain today that our proposed
ealism and applicability of our analysis, we introduced an ad
model is intended to fill in. Although current works have
ditional column named 'Price'. This new column allows us to
explored the utilization of KGs for structured data, there
consider the cost of medications, which is a significant factor
appears to be limited focus on the combination of both
in real-world drug recommendation systems..
structured and unstructured data from various sources [9],
[10], [11]. This way, our model takes advantage of a hybrid B. Dataset Overview
data collection approach, increasing the generalisability and
The dataset includes the following columns:
validity of sentiment analysis for the pharmaceutical
 Price: The cost of the drug, which was added to
industry. The evaluation metrics for our deep learning model
indicate its superior performance. With an accuracy of make the dataset more realistic and relevant for
89.92%, precision of 0.9208, recall of 0.9370, and an F1 practical applications.
score of 0.9288, our model outperforms traditional models  drugName: The name of the drug being reviewed.
that were tested on the same dataset. These metrics highlight  condition: The medical condition for which the
the model's robustness in correctly identifying positive and drug was prescribed.
negative sentiments, which is crucial for accurate drug  review: The text of the user review.
recommendations. On the contrary, these research works [1],  rating: The rating given by the user, ranging from 1
[2], [3] found low accuracy rather than ours. Many existing to 10.
models fall short of leveraging the capabilities of deep  date: The date the review was posted.
learning mechanisms for sentiment analysis, especially in  usefulCount: The number of users who found the
relation to drug reviews [12]. Our model uses complex deep review helpful.
learning structures that increase the clarity of the distinctions  sentiment: A derived column indicating the
made regarding sentiment. Another area of concern that has sentiment of the review (positive or negative),
not seen much research is the automation of drug based on the rating.
recommendations using sentiment analysis [13]. Our By incorporating the 'Price' column, we can better analyze
proposed model does not only predict the sentiments of the the cost-effectiveness of drugs alongside their effectiveness
brand but also utilizes these sentiments to recommend drugs, and user satisfaction. This addition makes our dataset more
which makes our model unique in comparison with other comprehensive and relevant for real-world applications,
models that are present in the literature. Most of the research
providing a more holistic approach to drug recommendation.
is centered around theoretical or backend processing,
The data used in this research work is combined, and the
disregarding immediate user involvement [15]. The built
model integrates a live question-and-answer session where sources are from the UCI Machine Learning Repository and
patients and healthcare providers can search for drug Kaggle. It includes user comments on different drugs and
recommendations for a particular condition, making the other additional information like the drugs used, conditions
study more useful for practice. A primary concern when they are intended for, conversion of overall ratings into
developing an ML model is the inherent bias that occurs average, dates of the reviews, and the number of useful
especially when training on user-generated content, and it is comments. Since medical information ranges from patients’
an issue that has not been fully explored in earlier works [14
emotions to drugs’ effectiveness, this versatile dataset lets
you study all the aspects of the population’s life.

Preprocessing is a critical step to ensure the quality and

consistency of the dataset. The following steps were
undertaken:
.
 Data Cleaning: Reviews were converted to strings,
and HTML tags were removed using
BeautifulSoup. Non-alphabetic characters were
filtered out using regular expressions.
 Stop Words Removal: Common English stop words
were removed to retain only meaningful words.
This was achieved using the NLTK library.
 Lowercasing: All text was converted to lowercase
to maintain uniformity. Fig.1. xwq
 N-grams Generation: N-grams (bi-grams and tri-
grams) were generated to capture the contextual E. Sequential Neural Network Model
information and enhance the sentiment analysis.
 Sentiment Labeling: Sentiments were labeled as Additionally, we developed a Sequential Neural
positive (1) for ratings above 5 and negative (0) for Network model with the following architecture:
ratings 5 or below.  Input Layer: Accepts the n-gram feature vectors.
 Dense Layer: A fully connected layer with 250
C. Vectorization neurons and ReLU activation function.
To further analyze the text data, we again utilized the scik  Dropout Layer: A dropout layer with a 20%
it-learn package to vectorize the predictions and the correspo dropout rate to prevent overfitting.
nding test set clean text data. The vectorizer was set to repeat  Output Layer: A single neuron with a sigmoid
n-gram features: This was set in a bid to a reduce the comput activation function to output the probability of the
ational power that the program would require to handle The p sentiment being positive.
arameter that was set to control the dimensionality of the vect The Sequential model was compiled using the Adam
or was the maximum feature number which was set at 5000. optimizer and binary cross-entropy loss function, and it was
Many of the subsequent steps involved in model training are trained over 5 epochs with a batch size of 128. The training
simplified when the matrix is in dense form, so the matrix wa process was monitored to ensure convergence and prevent
s converted to dense form. overfitting. The Sequential model achieved an accuracy of
D. Model Development 89.92%.
A deep learning model was developed using the Keras F. Enhancing of the Convolutional Neural Network (CNN)
library. The architecture of the model is as follows: Architecture
 Input Layer: Accepts the n-gram feature vectors. To increase the effectiveness of the deep learning model, we
 Dense Layer: A fully connected layer with 250 decided to integrate Convolutional Neural Network (CNN)
neurons and ReLU activation function. into the architecture; modifying it in a way that we can have
 Dropout Layer: A dropout layer with a 20% a better approach to dealing with text data particularly for
dropout rate to prevent overfitting. sentiment analysis. These enhancements and modifications
 Output Layer: A single neuron with a sigmoid of the CNN were made with the following factors in mind.
activation function to output the probability of the We included an embedding layer to transform the high-
sentiment being positive. dimensional sparse vectors, which are created by
The model was compiled using the Adam optimizer and CountVectorizer, into the low-dimensional dense vectors.
binary cross-entropy loss function, and it was trained over 5 This step is vital for extracting the meaning of words used in
epochs with a batch size of 128. The training process was the reviews. The layers of convolution were implemented
monitored to ensure convergence and prevent overfitting. with different sizes of filters to read different aspects of n-
grams and thus, it helps the model to learn from short
phrases and long sequences within the text to enhance the
ability to capture sentiment implications. Strategies of this
kind were meant to be placed on top of the text, and then the
convolution operation was performed, creating feature maps
that would point out certain patterns or phrases that
correspond to certain sentiments. To decrease the size of
feature maps and keep the most important information,
pooling layers (max pooling) were used. This step assists in
making the model numerically efficient; besides, it
minimizes the chance of overfitting the model. To further
reduce overfitting, we incorporated dropout after the
convolutional and dense layers. Dropout is a technique that
involves randomly ignoring a certain percentage of the input
units; this helps in making the model less sensitive to the Accuracy 89.92%
input. Also, batch normalization was applied to normalize
Precision 92.08%
the activation of each layer to speed up the training of the
model, which also has regularization effects. Recall 93.70%
F1 Score 92.88%
The optimization process used the Adam optimization
algorithm due to its efficiency and suitability when coping
with sparse gradients when the data is noisy. This optimizer
The model performance was then tested on a test set, equi
adapts the learning rate based on the training process to
valent to thirty percent of the total number of files in the data
increase the model’s convergence speed and its robustness. base. Newland defined the evaluation metrics that were used
ReLU (Rectified Linear Unit) activation functions were used in this research as accuracy, precision, recall, and the F1 scor
in the convolutional and dense layers to spice up the e. The model was able to predict the results up to a fairly goo
linearity and the sigmoid activation function was used in the d accuracy of 89 percent. Attaining a high accuracy of precisi
output layer to give the probability score for the binary on & F1 score 92%, hence arguing the fact that the proposed
sentiment classification. algorithm works effectively for sentiment classification. The
G. Model Evaluation detailed evaluation metrics are as follows: The detailed evalu
ation metrics are as follows:
The trained model was evaluated using a test set
comprising 30% of the original dataset. The evaluation
metrics included accuracy, precision, recall, and the
confusion matrix to provide a comprehensive assessment of
the model's performance.
H. Drug Recommendation System
These sentiments were used to create drug
recommendations for patients. For each drug, the system
then determines an average rating and average sentiment
score and then makes a composite ranking of the drugs. An
interface was developed where a form or an array of
questionnaires allows patients and physicians to search for
the given drug recommendations in concern with certain Fig.
diseases. From the ergonomics point of view, this interface
contributes to the model’s practical usability by offering These metrics demonstrate promising results, meaning
individualized tips in real time. that the model shows high accuracy in both positive and
negative emotions in drug reviews. The confusion matrix,
I. Addressing Bias depicted in the figure below, provides a visual
To mitigate biases in the model, several measures were representation of the model's predictions:The confusion
taken: matrix, depicted in the figure below, provides a visual
 Balanced Training Set: Ensuring the training set representation of the model's predictions:
had a balanced representation of positive and
negative sentiments.
 Cross-Validation: Using cross-validation
techniques to ensure the model's robustness across
different subsets of the data.
 Bias Detection: Implementing bias detection
mechanisms to identify and correct any skewed
predictions.
The identified approach, as shown in our methodology,
involves the incorporation of the complex NLP and deep
learning strategies into the framework of the Semantic Web
to improve the options of sentiment analysis and drug
recommendation. Due to the capacity of the model to test
the efficiency of the model on different datasets and the real-
time recommendations for users make the model very useful
in the area of health care.
IV. RESULTS
The results of our study demonstrate the effectiveness of Fig.
our deep learning model in accurately predicting sentiments f
rom drug reviews and providing relevant drug recommendati From the confusion matrix, it is evident that the model
ons based on these sentiments. Below are the detailed results performs well in distinguishing between positive and
of our model's performance and an example of the drug reco
negative sentiments. The true positive rate is significantly
mmendation output.
higher, which contributes to the overall high accuracy of the
Accuracy Matrics Score model.
A. Drug Recommendation Example 2020) eBay,
To demonstrate the practical utility of our model, we [17] Amazon
queried the system for drug recommendations for the and
condition "Depression." The model provided the following Flipkart
top five drug recommendations based on sentiment analysis: (Wu et E- E- BERT- 72.8%
al., commer commerc based
According to the analysis made in this study regarding the 2023) ce e data Sentiment
drugs for treating depression based on the composite score [16] Analysis
comprised of consumer rating and sentiment, the table (Shar- Movie Movie SPRS Avg
below reveals the five most recommended drugs. All of ma et Site Site Data (Deep Accurac
these medications got the overall rating of 100%, which al., Learning y: 61.5
demonstrates that users have a high level of satisfaction with 2021) Base
these drugs, as well as positive attitudes. [18] Model)
Drug name Score Rating Sentiment (Nguy Movie Movie Tag Avg
en et Site Site Data Interpolatio Accurac
Asendin 10 10 1 al., n y: 58.7
Niravam 10 10 1 2020)
[19]
Norpramin 10 10 1 (Rawat Movie MovieLe EMBEDDI MAE:
Maprotiline 9 9 1 et al., Site ns: Movie NG + ANN 0.7305
2020) Data &
Xanax XR 8 8 1 [20] RMSE:
Table 0.9311
Table
B. Compared with other Models We implemented and evaluated several other machine
For the purpose of strengthening our results, we first com learning models, including LightGBM, XGBoost, LSTM,
pared the outcomes of the deep learning model proposed with Multinomial Naive Bayes (MNB), and Random Forest, on
the LightGBM and XGBoost and with the LSTM, MNB, and the same dataset. Despite their strengths, these models
Random Forest. But on comparing these models with our dee showed lower performance metrics compared to our deep
p learning models none of them proved fruitful as the propos learning model. For instance, LightGBM and XGBoost
ed deep learning model. The performance metrics for these m struggled with the imbalanced nature of the dataset, leading
odels are summarized below:The performance metrics for the to lower precision and recall rates. This demonstrates the
se models are summarized below: effectiveness of our custom approach in capturing the key
Model Accuracy features of the text data, leading to better sentiment
LightGBM 82.04% prediction.
Random Forest 70.33%
XGBoost 65.96% V. CONCLUSION
LSTM 54.70% Our study demonstrates the efficacy of a deep learning
Table model integrated within the Semantic Web framework for
sentiment analysis and drug recommendation. Achieving an
accuracy of 89.92%, with high precision, recall, and F1
C. Compare with Conventional CNN Model score, the model effectively interprets user sentiments from
In our study, we compared the performance of a custom drug reviews and provides reliable recommendations. The
model against a conventional Convolutional Neural Network model's ability to generalize across diverse datasets and its
(CNN) model. The custom model was specifically designed interactive user interface underscore its potential to enhance
to handle the nuances of text data, while the conventional personalized healthcare. By addressing existing research
CNN was implemented with a standard approach to process gaps, our approach offers a robust solution for real-time,
the text as a sequence of words. sentiment-based drug recommendations, paving the way for
Model Accuracy (%) improved patient outcomes and satisfaction.
Custom Model 89.92
Conventional CNN 88.40 REFERENCES
Table
[1] P. Karmalkar, H. Gurulingappa, E. Spies, and J. A. Flynn, "Artificial
The custom model achieved an accuracy of 89.92%, intelligence-driven approach for patient-focused drug development," Front.
Artif. Intell., vol. 6, p. 1237124, Oct. 2023. DOI:
outperforming the conventional CNN model, which
10.3389/frai.2023.1237124.
provided an accuracy of 88.40%.
[2] M. Kulmanov, F. Z. Smaili, X. Gao, and R. Hoehndorf, "Semantic
Articl- Field Data Model/ Accura similarity and machine learning with ontologies," Briefings in
Bioinformatics, vol. 22, no. 4, pp. bbaa199, 2021.
e Type Algorithm cy
Type [3] S. K. Narayanasamy, K. Srinivasan, Y.-C. Hu, S. Masilamani, and K.-
(Shar- E- E- Heuristic Avg Y. Huang, "A Contemporary Review on Utilizing Semantic Web
ma & commer commerc Algorithm Accurac Technologies in Healthcare, Virtual Communities, and Ontology-Based
Information Processing Systems," Electronics, 2022.
Rana, ce e data: y: 53.2
[4] Y. Chudasama, D. Purohit, P. D. Rohde, and M.-E. Vidal, "Enhancing [13] A. A. Kumar, "Semantic memory: A review of methods, models, and
Interpretability of Machine Learning Models over Knowledge Graphs," in current challenges," Psychonomic Bulletin & Review, vol. 28, no. 1, pp.
International Conference on Semantic Systems, 2023. [Online]. Available: 40-80, 2021.
https://api.semanticscholar.org/CorpusID:265309633
[5] T. Lertchaturaporn and P. Songmuang, "Automated Scoring Model for [14] Y. Y. Tan, C. O. Chow, J. Kanesan, J. H. Chuah, and Y. Lim,
Semi-Open-Ended Question in Scientific Explanation," in 2024 IEEE "Sentiment Analysis and Sarcasm Detection using Deep Multi-Task
International Conference on Cybernetics and Innovations (ICCI), 2024, pp. Learning," Wireless Personal Communications, vol. 129, pp. 2213-2237,
1-5. 2023. [Online]. Available:
https://api.semanticscholar.org/CorpusID:257355920
[6] S. Ouyang, "Deep learning for sentiment analysis on IMDB movie
reviews using N-gram features," Applied and Computational Engineering, [15] C. Prasitpuriprecha et al., "Drug-Resistant Tuberculosis Treatment
2024. [Online]. Available: Recommendation, and Multi-Class Tuberculosis Detection and
https://api.semanticscholar.org/CorpusID:267442287 Classification Using Ensemble Deep Learning-Based System,"
Pharmaceuticals, vol. 16, 2022. [Online]. Available:
[7] B. Kotiyal, H. Pathak, and N. Singh, "Debunking multi-lingual social https://api.semanticscholar.org/CorpusID:255092681
media posts using deep learning," International Journal of Information
Technology, vol. 2023, pp. 1-13. [Online]. Available: [16] Wu HC, Luk RWP, Wong KF, Kwok KL. (2023). Sentiment analysis
https://api.semanticscholar.org/CorpusID:259082609 using deep learning techniques: a comprehensive review. International
Journal of Multimedia Information Retrieval.
[8] Lynda, D., Brahim, F., Hamid, S., & Hamadoun, C. (2023). Towards a
semantic structure for classifying IoT agriculture sensor datasets: An [17] S. Sharma and V. Rana, "Web Search Personalization Using Semantic
approach based on machine learning and web semantic technologies. Similarity Measure," in Proceedings of ICRIC 2019, P. K. Singh, A. K.
Journal of King Saud University-Computer and Information Sciences, Kar, Y. Singh, M. H. Kolekar, and S. Tanwar, Eds. Cham: Springer
35(8), 101700. International Publishing, 2020, pp. 273–288. doi.org/10.1007/978-3-030-
29407-6_21
[9] Suzuki, Y., & Elgort, I. (2024). Measuring automaticity in second
language comprehension: A methodological synthesis of experimental tasks [18] S. Sharma, V. Rana, and V. Kumar, "Deep learning based semantic
over three decades (1990–2021). Practice and Automatization in Second personalized recommendation system," International Journal of Information
Language Research: Perspectives from Skill Acquisition Theory and Management Data Insights, vol. 1, no. 2, p. 100028, 2021. DOI:
Cognitive Psychology. New York: Routledge. 10.1016/j.jjimei.2021.100028.

[10] Hafidi, M. M., Djezzar, M., Hemam, M., Amara, F. Z., & Maimour, [19] Q. N. Nguyen, N. Duong-Trung, D. N. L. Ha, X. S. Ha, T. T. Phan, H.
M. (2023). Semantic web and machine learning techniques addressing X. Pham, and H. X. Huynh, "Movie Recommender Systems Made Through
semantic interoperability in Industry 4.0. International Journal of Web Tag Interpolation," in Proceedings of the 4th International Conference on
Information Systems, 19(3/4), 157-172. Machine Learning and Soft Computing, Haiphong City, Viet Nam, 2020,
pp. 154–158, doi: 10.1145/3380688.3380712.
[11] M. Yahya, J. G. Breslin, and M. I. Ali, "Semantic web and knowledge [20] R. M. Rawat, V. Tomar and V. Kumar, "An Embedding-based Deep
graphs for industry 4.0," Applied Sciences, vol. 11, no. 11, p. 5110, 2021. Learning Approach for Movie Recommendation," 2020 5th International
Conference on Communication and Electronics Systems (ICCES),
[12] P. Hitzler, "A review of the semantic web field," Communications of Coimbatore, India, 2020, pp. 1145-1150, doi:
the ACM, vol. 64, no. 2, pp. 76-83, 2021. 10.1109/ICCES48766.2020.9137998.