Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886

Comparative Analysis of Large Language

Models and Traditional Methods for
Sentiment Analysis of Tweets Dataset
Suhag Pandya

Abstract:- Sentiment analysis is widely recognised as the Sentiment analysis (SA) informs consumers before they
most actively researched area in data mining. These days, purchase a product whether the information is suitable [6][7].
a number of social media platforms have been created, and This analytical data is used by marketers and businesses to
Twitter is a crucial instrument for exchanging and better understand their goods and services so that they may be
gathering people's thoughts, feelings, opinions, and provided according to the user's requirements [8][9].
attitudes about specific things. This paper presents a Processing, finding, or analysing the factual material available
comparative analysis of Large Language Models (LLMs) is the major emphasis of textual information retrieval strategies
and ML models for sentiment analysis on a Twitter [10].
dataset. The study evaluates a performance of XLNet,
advance algorithms including KNN, RF, and XGBoost. Worldwide, thanks to technological advancements, social
The sentiment analysis methodology involves pre- media platforms like Instagram, Facebook, LinkedIn,
processing the Twitter dataset through noise removal and YouTube, etc. [11][12]. people's ideas and thoughts on things,
tokenisation, followed by feature extraction using methods events, or objectives can be expressed through [13][14]. In
like Bag-of-Words and Word2Vec. Results show that today's world, people all over the world love to communicate
XLNet is superior to the conventional models; It has 99% their opinions and thoughts through short messages called
precision, recall, and F1-score values and an accuracy rate tweets on the popular microblogging network known as Twitter
of 99.54%. In comparison, KNN achieves 78% accuracy, [15][16]. An enormous amount of sentiment data derived from
85% precision, 88% recall, and 86% F1 score, while RF analyses of tweets is frequently generated [17]. Twitter is a
and XGBoost exhibit lower performance with accuracy great platform for sharing news and interacting with other users
rates of 69% and 60%, respectively. The performance online [18][19]. The way people feel on Twitter has a big
comparison highlights the superior capabilities of XLNet impact on many parts of our lives. The goal of SA and text
for sentiment classification tasks, indicating its potential classification is to extract information from texts and then
for enhancing text classification applications. Research in classify a polarity as neutral (Ne), negative (N), or positive
the future can look at ways to improve XLNet's (P)[20].
performance on bigger and more complicated datasets by
combining it with more sophisticated deep learning Traditional machine learning (ML) approaches have long
methods like attention mechanisms and transfer learning. been utilised for SA tasks, relying on extensive preprocessing
and feature extraction techniques [21][22]. However, the
Keywords:- Sentiment Analysis, LLM, Twitter Dataset, advent of LLMs such as GPT and BERT has transformed the
Natural Language Processing, Text Classification, Social landscape of NLP [23][24]. These models, pre-trained on vast
Media Analytics. corpora of text, demonstrate a superior ability to comprehend
context, capture nuances, and generate accurate predictions,
I. INTRODUCTION thereby outperforming conventional methods in many
scenarios[25][26]. The main objective of this project is to use
The Internet has revolutionised the way individuals share Twitter datasets to evaluate and contrast LLMs with more
their thoughts and ideas these days. These days, most people do traditional ML techniques for sentiment analysis.[27]. By
it on websites that allow product reviews, social media, online leveraging the strengths of both paradigms, this research seeks
forums, and blogs [1]. Facebook, Twitter, Google Plus, and to highlight their efficacy, limitations, and practical
many other social media platforms are attracting millions of applications.
users every day [2]. To share their thoughts and feelings as well
as comments regarding their everyday lives [3]. Consumers  Significance and Contribution of Study
may educate and persuade one another through online forums This study highlights a significance of leveraging both
and online communities, which is an interactive kind of media LLMs and traditional methods for sentiment analysis to
[4]. Blog entries, reviews, comments, tweets, and status enhance accuracy and contextual understanding. Finding out
updates all contribute to the mountain of sentiment-rich data how well each method worked on a real-world Twitter dataset
that is generated by social media [5]. allows us to compare and contrast their merits and
shortcomings. The findings contribute to advancing sentiment
analysis techniques, enabling more effective applications in
social media analytics and NLP. The contribution of this study

IJISRT24DEC1694 www.ijisrt.com 1647

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886
lies in its comprehensive evaluation of various Large language In this study, Gore et al. (2023) conducted preprocessing
models for sentiment analysis using the Twitter dataset. The using tokenisation, stemming, punctuation mark removal, and
main contributions are: stop word removal. Word clouds may be used for both positive
and negative word analysis. They got both good and bad
 Analysed a substantial Twitter dataset with over 31,000 comments on the Apple iPhone 14 Pro Max. Reviews that are
tweets, ensuring the study reflects practical applications. favourable tend to be ones that are dynamic, good, improving,
 Captured semantic relationships by transforming words gorgeous, terrific, excellent, etc. Several machine learning
into vectors, enabling context-aware sentiment analysis. techniques were used, including NB, LR, and SVM.
 Identified key sentiment-related features, enhancing model LR achieved 90% accuracy, SVM 91% accuracy, and NB 94%
efficiency and predictive accuracy. accuracy [30].
 Provided a foundational baseline for sentiment analysis,
enabling comparison with advanced techniques like In this study, Waspodo et al. (2022) The steps that
Word2Vec. makeup preprocessing include cleaning, lemmatisation,
 Compare the effectiveness of LLMs, such as XLNet, with stemming, tokenising, and stop word removal. This study
traditional ML models like RF, XGBoost, and KNN for employs a variety of unsupervised learning techniques:
sentiment analysis of Twitter data. Supervised learning based on Lexicon: SVM. The normalised
 The models were evaluated using a number of relevant Lexicon Sentiment with scaling is used to establish positive,
measures, including F1-score, recall, accuracy, and neutral, and negative sentiment classes, and this data is then
precision. used to weight textual data. With 940 tweets analysed, they
found 330 positive (35% of the total), 302 negative (32% of the
 Structure of the Paper total), and 308 neutral (33% of the total) tweets. The findings
The research is organised as follows: In Section II, the showed that the SVM achieved an average accuracy of 98.75%
existing literature on text classification is reviewed. In section in its classification tests, with a precision of 0.98 and a recall of
III Methodology utilised to compile the data for this study. The 0.98 [31].
findings and commentary on the text categorisation are
presented in Section IV Section V concludes with a present In this study, Asmawati, Saikhu and Siahaan (2022) look
finding. at four different sentiment analysis approaches and evaluate
their efficacy on text and image-based Indonesian memes. NB,
II. LITERATURE REVIEW SVM, DT, and CNNs were the supervised ML algorithms used
to classify the retrieved text memes after their extraction. The
A survey of the literature on sentiment is included in this study results showed that the most beneficial effects were
section. Text categorisation using analysis of the Tweets obtained using sentiment analysis on meme text using the
dataset. NB technique, with an accuracy rate 65.4% [32].

In this study, Ihnaini et al. (2024) fine-tuning process In this study, Gope et al. (2022) Sentiment analysis seeks
incorporates both supervised techniques and reinforcement to categorise an amount of good and negative emotions
learning from human feedback, specifically designed to align expressed in a certain text. ML methods such as LR, MNB,
the models with the historical and cultural context of Song Ci. BNB, RF, and LSVM were used. They achieved a 91.90%
Notably, the ChatGLM-6B(8-bit) model achieved the best F1 success rate using the RFC. In our study, they also achieved
Score of 0.840, demonstrating its exceptional ability to merge maximum accuracy (97.52%) using RNN with LSTM as a
ancient literary analysis with modern computational DL strategy. Our model is well-suited to the RNN-LSTM
technology, thereby broadening our understanding of the technique [33].
emotional spectrum of classical Chinese poetry [28].
In this paper, Al-Hagree and Al-Gaphari (2022) Data used
In this study, Patrick et al. (2023) The model is developed for SA tasks were collected from evaluations left by users of
and tested using a variety of learning algorithms, including banking mobile applications on the Google Play Store. The
Linear Regression, KNN, SVM, RF, Bagging, and Gradient Arabic sentiment analysis is carried out by use of ML
Boosting. There are two sets of data: the test dataset and the algorithms, namely the NB, KNN, DT, and SVM patterns. NB
training dataset. The test as well as training datasets were used model has outperformed competing DT, KNN, and SVM
to create and evaluate the model. It was determined that 1.1% algorithms in terms of evaluation quality. In comparison to the
of respondents were untrustworthy due to unanswered other models, the NB model had exceptional results according
questions by the reliability check. The results demonstrate that to recall(88.08%), accuracy(88.25%), and F-score(88.25%)
out of the two algorithms, RF-supervised ML yields a superior [34].
accuracy of 0.711 compared to KNN's 0.515 [29].
Below, Table 1 provides a summary of a literature review
with dataset approaches, results and limitations for text dataset
classification.

IJISRT24DEC1694 www.ijisrt.com 1648

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886
Table 1 Summary of Literature Review of Sentiment Analysis of Tweets Dataset for Text Classification
Author Methods Dataset Accuracy Limitation/gap
Patrick Linear Regression, KNN, Dataset of 884 Random Forest: 71.1%, Limited reliability control, only
et.al. SVM, RF, Bagging, GB. employees was KNN: 51.5% 1.1% deemed unreliable; lower
reduced to 802 after accuracy with certain models
cleaning
Gore et al. Naive Bayes, SVM, Reviews on Apple Logistic Regression: 90%, No mention of deep learning or
Logistic Regression iPhone 14 Pro Max SVM: 91%, Naive Bayes: ensemble techniques
94%
Ihnaini ChatGLM-6B (8-bit) Ancient Chinese F1 Score: 0.840 Focused on a very niche area,
et.al. model poetry difficult to generalise results to
modern language sentiment
analysis
Waspodo Lexicon-based 940 tweets SVM: Accuracy: 98.75%, Narrow focus on tweets, results
et.al. (unsupervised) & SVM recall: 0.98, precision: might not generalise to other
(supervised) 0.98 text types.
Asmawati, Naive Bayes, SVM, Indonesian memes Naive Bayes: 65.4% Accuracy lower for complex
Saikhu Decision Tree, (text and images) text and image data
et.al. Convolutional Neural combinations
Networks
Gope et.al. Linear SVM, RF, Naive Amazon product Random Forest: 91.90%, Limited focus on deep learning
Bayes (Multinomial & reviews RNN-LSTM: 97.52% methods apart from RNN-
Bernoulli), LR, RNN- LSTM
LSTM
Al-Hagree Naive Bayes, KNN, Mobile app user Naive Bayes: 89.65%, Focused only on Arabic
and Al- Decision Tree, SVM reviews (Arabic) recall: 88.08%, precision: language, other language
Gaphari et 88.25%, F1: 88.25% models not explored.
al.

III. METHODOLOGY

The methodology for the comparative analysis of Large

Language Models (LLMs) and traditional methods for
sentiment analysis of the Twitter dataset involves several key
steps. Figure 1 shows a workflow of flowchart sentiment
analysis for text classification employing ML models to
analyse Twitter data. The sentiment analysis methodology
involves analysing a Twitter dataset labelled with Positive and
Negative sentiments. The dataset undergoes pre-processing to
clean and standardise text, including noise removal and
tokenisation. Traditional methods like Bag-of-Words and
Word2Vec are used for feature extraction, capturing
sentiment-relevant features such as word frequencies and
vector similarities. The processed For model assessment, data
is divided into training (80%) and testing (20%) sets. Both
Large Language Models (LLMs) like XLNet and traditional
ML algorithms like RF, XGBoost, and KNN are applied for
classification. The models are trained and tested, and The F1-
score, accuracy, precision, and recall are the typical measures
used to assess their performance. Finally, the models forecast
a sentiment of tweets as either Positive or negative, and their
results are compared to determine the most effective approach
for text classification.

Fig 1 Flowchart for Sentiment Analysis

IJISRT24DEC1694 www.ijisrt.com 1649

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886
A. A following Steps of flowchart are Explained below:  Removing URLs: Eliminate any hash marks (like
#subject), URLs (like www.xyz.com), and targets
 Dataset Description (@username)[39].
Use the developer account to gather the Twitter API for  Remove HTML: As a text preparation step, HTML tag
this investigation. In training datasets, there are 31962 tweets removal helps clean up HTML documents' text data
in total, with "29959" representing positive values and "2003" [40][41].
representing negative values. There were four categories of  Stemming: The stemming technique is used to remove
perplexity in the data they gathered: joyful, sad, cheerful, and prefixes and suffixes from words [42]. The process of
hateful. They concluded that the data had to be in binary form identifying and eliminating a word's root and stem is
once they organised it [35]. For this, they made cheerful and another possible definition [43].
cheery “Positive” and sad and hate “Negative”. In the realm  Tokenization: The process of tokenisation involves
of data management, this exists [36]. separating text into a set of discrete, meaningful
components [44].
 Data Preprocessing
After the dataset has been collected, the following stage  Bag-of-Words
is to preprocess it in order to eliminate any junk values and The information can be easily transformed into a bag-of-
extraneous information that may have been overlooked since words format [45][46]. In this representation, some common
it is a real-life dataset [37][38]. The following preprocessing words, like ‘go’ and ‘see,’ show a high distinction between
steps are listed below: classes, while others occur in similar amounts for every class.

Fig 2 Most Common Words across Sentiments

Figure 2 illustrates sentiment analysis across common enhanced by feature extraction [49]. Emotion counts (both
words, categorised into positive (blue), negative (orange), and positive and negative), question marks, hashtags, and
neutral (green) sentiments. Words like "thi," "go," "time," exclamation points are some of the key characteristics [50].
"like," and names like "trump" and "obama" are analysed, with
the y-axis showing frequency counts up to 250. This analysis,  Word2Vec
likely derived from social media or text data, reveals how The process of word2vec involves converting words to
word usage correlates with sentiment expression, offering vectors and then identifying words that are similar to each
insights into linguistic and emotional trends in the dataset. other. This enables it to identify words that reflect distinct
emotions [51]. The construction of Word2Vec involves
 Feature Extraction calculating the average similarity of a whole tweet to a given
Feature extraction is a powerful tool for reducing word, with the average values normalised to [0,1] to account
resource requirements while preserving critical data [47][48]. for variations in word count across tweets [52][53]. Figure 3
The effectiveness and precision of ML models are greatly displays the terms that are comparable.

IJISRT24DEC1694 www.ijisrt.com 1650

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886

Fig 3 Words Similarity

Figure 3 visualises word similarity patterns across three 𝑇

keywords - "good," "bad," and "information" - showing their max 𝐸𝑧 ~𝑍𝑇 { ∑ logp(𝑥𝑧𝑡 |𝑥𝑧 < 𝑡)} (1)
𝑡=1
associations with different sentiment categories. For each
word, the bars display varying heights indicating the Where 𝑧𝑡 and z < t represent a permutation's t-th and t-1
frequency or strength of association, with "bad" showing the elements, respectively. 𝑍𝑇
highest positive sentiment count (around 1000), followed by
"good" (around 800), while "information" has more balanced A likelihood of token 𝑥𝑧𝑡 given previous tokens 𝑥𝑧𝑡 <𝑡 is
sentiment distributions across all three categories. This computed using the XLNet auto-regressive permutation
visualisation effectively demonstrates how words that might approach, as displayed in Equation 1. Although XLNetonly
seem inherently positive or negative can actually have modifies the factorisation order and not a sequence_order
complex sentiment associations in real-world usage. during training, it preserves an original sequence_order and
utilises Transformers to match the original sequence's
 Data Splitting positional encoding[62][63]. This quality is helpful for
Data splitting in ML models is a crucial step for fortuning as it takes into account just the sequence's natural
evaluating the performance of models. The basic premise is to order [64]. They thus use XLNet in our research since its
partition the dataset into separate parts for testing and training. design differs from that of BERT [65][66].
They use 80% of the data for training and just 20% for testing.
 Performance Metrics
 Large Language Model (XLNET) To evaluate the performance of phishing email detection,
XLNet[54] is one of Google AI's 2019 Transfer Learning used a set of evaluation metrics, also known as performance
models; it's similar to BERT but uses an AR pre-training metrics [67][68]. A confusion matrix may be used to compare
strategy to generalise its features, making it perform better the predicted and actual values of a model in order to evaluate
than BERT on many benchmark datasets [12][55]. They its accuracy [69][70]. Five measures were utilised to evaluate
demonstrate below how XLNet has used PLM to overcome the final models: accuracy, precision, recall, and score. This
AE models' shortcomings, namely the issue of obtaining evaluation process begins with confusion matrices that use the
bidirectional context [56]. In comparison to BERT, XLNet's following criteria to rank the models: with TP, FP, TN, and
convergence time is much longer due to the fact that it trains FN, where:
through all possible word sequences utilising permutations of
occurrences for a particular word [57][58].  TP indicates the amount of occurrences when the actual
class was accurately predicted.
The core concept of XLNet is to enhance PLM with  FN, the picture shows how many times the real class was
additional capabilities in order to capture situations that are wrongly assumed to be some other kind.
bidirectional [59][60]. It is possible to accomplish AR
 TN displays the quantity of correctly classified records as
factorisation in T! by thinking about every possible location
normal.
on each side of a token. separate sequences x tokens of length
 FP, it is the sum of all the times a different class was
T in a phrase [61]. Consider ZT to be the collection of every
wrongly assumed to be the real class.
combination of T-length sequences that could exist.

IJISRT24DEC1694 www.ijisrt.com 1651

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886
 Accuracy: The ratio of properly categorised undesirable  F1 Score: The following Eq. (5) is the formula for the F1
and normal events to total oil well events in the dataset is Score, which is an all-encompassing assessment that
known as accuracy. It is stated as follows Eq.(2): strikes a balance between recall and precision values:
𝑇𝑃+𝑇𝑁 2×𝑟𝑒𝑐𝑎𝑙𝑙×𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = 𝑇𝑃+𝐹𝑃+𝐹𝑁+𝑇𝑁 (2) 𝐹1 = (5)
𝑟𝑒𝑐𝑎𝑙𝑙+𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛

 Precision: According to formula (3), precision is a The purpose of these matrices is to facilitate the
percentage of positive observations that accurately evaluation of various ML models on the Twitter dataset.
forecast a total number of positive predictions.
IV. RESULT ANALYSIS AND DISCUSSION
𝑇𝑃
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = (3)
𝑇𝑃+𝐹𝑅
The experiment is done on the twitter dataset. On this
twitter data apply Large Language Model (XLNET), and
 Recall: Recall is a percentage of properly detected positive
compare (see in Table 3) with KNN[71], Random Forest[72],
observations, as computed using Eq. (4):
XGBoost [73] models. Table II displays the outcomes of a
𝑇𝑃 sentiment analysis using a LLM model.
𝑅𝑒𝑐𝑎𝑙𝑙 = (4)
𝑇𝑃+𝐹𝑁

Table 2 XLNet model Performance on Twitter Dataset

Measure XL Net
Accuracy 99.54
Precision 99
Recall 99
F1-score 99

Fig 4 Performance of XL-Net Model

The Figure 4 bar graph displayed a performance of a XLNet model for Twitter data. In this graph, the XLNet model's
performance for sentiment analysis achieves an accuracy of 99.54, precision of 99%, recall of 99%, and F1-Score of 99%, reflecting
its excellent overall results in classification tasks.

IJISRT24DEC1694 www.ijisrt.com 1652

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886

Fig 5 Accuracy Plot for XLNet Model

Figure 5 is a line plot showing the model accuracy of XLNet during training and validation across epochs. The accuracy is
displays on a y-axis, while a number of epochs is displayed on an x-axis. A plot shows a fluctuation in validation accuracy around
epochs 2-3, with a notable dip, while training accuracy remains relatively stable.

Fig 6 Loss Plot for XLNet Model

The following Figure 6 shows the loss plot for XLNet during training. An x-axis spans epochs between 0 and 4, and the y-axis
represents loss values from approximately 0.01 to 0.03. This graph is useful for tracking the model's learning progress and seeing
signs of under- or overfitting.

Table 3 Comparison between LLM and Another ML models Performance

Measures XLNet KNN RF XGBoost
Accuracy 99.54 78 69 60
Precision 99 85 44 56.3
Recall 99 88 88 51.3
F1-score 99 86 59 52.3

IJISRT24DEC1694 www.ijisrt.com 1653

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886
The following Table 3 presents the performance of [4]. J. Thomas, “A Survey of E-Commerce Integration in
experiments that used ML for text classification on the Twitter Supply Chain Management for Retail and Consumer
dataset. A number of metrics were used to assess an efficacy Goods in Emerging Markets,” J. Emerg. Technol.
of a RF, XGBoost, XLNet, and KNN models, including recall, Innov. Res., vol. 10, no. 12, 2023.
accuracy, precision, and F1-score. In this comparison, KNN [5]. V. A. and S. S. Sonawane, “Sentiment Analysis of
achieving superior results with accuracy78%, precision85%, Twitter Data: A Survey of Techniques,” Int. J.
recall88%, and F1-score86%, Random Forest achieves 69% Comput. Appl., 2016, doi: 10.5120/ijca2016908625.
accuracy, with 44% precision, 88% recall and a 59% F1-score [6]. Sahil Arora and Pranav Khare, “AI/ML-Enabled
and XGBoost achieves 60% accuracy with 56.3% of Optimization of Edge Infrastructure: Enhancing
precision, 51.3 of recall and 52.3% of F1-score indicating Performance and Security,” Int. J. Adv. Res. Sci.
relatively moderate performance and XLNet is performed Commun. Technol., vol. 6, no. 1, pp. 1046–1053,
strongly, with an accuracy 99.54%, precision 99%, recall 2024, doi: 10.48175/568.
99%, and F1-score 99%. Overall, the XLNet achieves the [7]. Pranav Khare and Shristi Srivastava, “Data-driven
highest performance in contrast with other models. product marketing strategies: An in-depth analysis of
machine learning applications,” Int. J. Sci. Res. Arch.,
V. CONCLUSION AND FUTURE WORK vol. 10, no. 2, pp. 1185–1197, Dec. 2023, doi:
10.30574/ijsra.2023.10.2.0933.
As a result of technological advancements, social media [8]. K. Patel, “AN ANALYSIS OF QUALITY
has become a vital component of people's daily lives. A ASSURANCE FOR BUSINESS INTELLIGENCE
person's innermost feelings, beliefs, and opinions may be PROCESS IN EDUCATION SECTOR,” IJNRD - Int.
easily shared via social media. Sentiment analysis's objective J. Nov. Res. Dev., vol. 9, no. 9, pp. a884–a896, 2024.
is to find out how a piece of writing is generally felt by [9]. V. S. Thokala, “Improving Data Security and Privacy
analysing whether it's good, negative, neutral, or has some in Web Applications : A Study of Serverless
other subjective feeling. The significance of sentiment Architecture,” Int. Res. J., vol. 11, no. 12, pp. 74–82,
analysis on strategic decision-making makes it crucial for both 2024.
business and society. In summary, XLNet performs noticeably [10]. S. Elmitwalli and J. Mehegan, “Sentiment analysis of
better than KNN, RF, and XGBoost when compared to COP9-related tweets: a comparative study of pre-
traditional ML models for sentiment analysis on the Twitter trained models and traditional techniques,” Front. Big
dataset. It achieves an impressive 99.54% accuracy rate along Data, vol. 7, 2024, doi: 10.3389/fdata.2024.1357926.
with high precision, recall, and F1-score. This research does [11]. S. G. Thomas Jubin, Kirti Vinod Vedi, “Enhancing
have some limitations, however. One of them is that it only Supply Chain Resilience Through Cloud-Based SCM
used one dataset, which could not be representative of the and Advanced Machine Learning: A Case Study of
complexity of actual sentiment analysis jobs in business. Logistics,” J. Emerg. Technol. Innov. Res., vol. 8, no.
Future work can focus on applying XLNet to larger, more 9, 2021.
diverse datasets, including multilingual text, to assess its [12]. N. Abid, “A Climbing Artificial Intelligence for
generalizability. Improving the model's performance and its Threat Identification in Critical Infrastructure Cyber
suitability for large-scale sentiment analysis applications in Security,” Int. J. Res. Anal. Rev., vol. 9, no. 4, 2022.
real time might be achieved by integrating Twitter data and [13]. S. S. Pranav Khare, “The Impact of AI on Product
investigating advanced approaches like hyperparameter Management : A Systematic Review and Future
optimisation and transfer learning. Trends,” Int. J. Res. Anal. Rev. (IJRAR, vol. 9, no. 4,
2022.
REFERENCES [14]. A. P. A. Singh and N. Gameti, “Leveraging Digital
Twins for Predictive Maintenance : Techniques ,
[1]. B. Patel, V. K. Yarlagadda, N. Dhameliya, K. Challenges , and Application,” IJSART, vol. 10, no. 9,
Mullangi, and S. C. R. Vennapusa, “Advancements in 2024.
5G Technology: Enhancing Connectivity and [15]. V. V. Kumar, F. W. Liou, S. N. Balakrishnan, and V.
Performance in Communication Engineering,” Eng. Kumar, “Economical impact of RFID implementation
Int., vol. 10, no. 2, pp. 117–130, 2022, doi: in remanufacturing: a Chaos-based Interactive
10.18034/ei.v10i2.715. Artificial Bee Colony approach,” J. Intell. Manuf., vol.
[2]. S. Bauskar and S. Clarita, “AN ANALYSIS: EARLY 26, no. 4, pp. 815–830, Aug. 2015, doi:
DIAGNOSIS AND CLASSIFICATION OF 10.1007/s10845-013-0836-9.
PARKINSON’S DISEASE USING MACHINE [16]. S. G. Jubin Thomas, Kirti Vinod Vedi, “Effects of
LEARNING TECHNIQUES,” Int. J. Comput. Eng. supply chain management strategies on the overall
Technol., vol. 12, no. 01, pp. 54-66., 2021, doi: performance of the organization,” Int. J. Sci. Res.
10.5281/zenodo.13836264. Arch., vol. 13, no. 01, pp. 709–719, 2024.
[3]. V. V. Kumar, A. Sahoo, S. K. Balasubramanian, and [17]. S. Bauskar, “A Review on Database Security
S. Gholston, “Mitigating healthcare supply chain Challenges in Cloud Computing Environment,” Int. J.
challenges under disaster conditions: a holistic AI- Comput. Eng. Technol., vol. 15, pp. 842–852, 2024,
based analysis of social media data,” Int. J. Prod. Res., doi: 10.5281/zenodo.13922361.
2024, doi: 10.1080/00207543.2024.2316884.

IJISRT24DEC1694 www.ijisrt.com 1654

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886
[18]. K. Patel, “The Impact of Data Quality Assurance [30]. R. D. Gore, V. Y. Borole, R. S. Gupta, V. Sonaje, P.
Practices in Internet of Things (IoT) Technology,” Int. Chaudhari, and B. W. Gawali, “Sentiments Analysis
J. Tech. Innov. Mod. Eng. Sci., vol. 10, no. 10, pp. 1– on Amazon Product Reviews Using Supervised
8, 2024. Machine Learning Algorithms,” in 2023 International
[19]. V. S. Thokala, “Integrating Machine Learning into Conference on Integration of Computational
Web Applications for Personalized Content Delivery Intelligent System, ICICIS 2023, 2023. doi:
using Python,” Int. J. Curr. Eng. Technol., vol. 11, no. 10.1109/ICICIS56802.2023.10430298.
6, pp. 652–660, 2021, doi: [31]. B. Waspodo, Q. Aini, F. R. Singgih, R. H.
https://doi.org/10.14741/ijcet/v.11.6.9. Kusumaningtyas, and E. Fetrina, “Support Vector
[20]. N. Parveen, P. Chakrabarti, B. T. Hung, and A. Shaik, Machine and Lexicon based Sentiment Analysis on
“Twitter sentiment analysis using hybrid gated Kartu Prakerja (Indonesia Pre-Employment Cards
attention recurrent network,” J. Big Data, 2023, doi: Government Initiatives),” in 2022 10th International
10.1186/s40537-023-00726-3. Conference on Cyber and IT Service Management,
[21]. S. K. R. Anumandla, V. K. Yarlagadda, S. C. R. CITSM 2022, 2022. doi:
Vennapusa, and K. R. V Kothapalli, “Unveiling the 10.1109/CITSM56380.2022.9935990.
Influence of Artificial Intelligence on Resource [32]. E. Asmawati, A. Saikhu, and D. Siahaan, “Sentiment
Management and Sustainable Development: A Analysis of Text Memes: A Comparison among
Comprehensive Investigation,” Technol. \& Manag. Supervised Machine Learning Methods,” in
Rev., vol. 5, no. 1, pp. 45–65, 2020. International Conference on Electrical Engineering,
[22]. Abhishek Goyal, “Driving Continuous Improvement Computer Science and Informatics (EECSI), 2022.
in Engineering Projects with AI-Enhanced Agile doi: 10.23919/EECSI56542.2022.9946506.
Testing and Machine Learning,” Int. J. Adv. Res. Sci. [33]. J. C. Gope, T. Tabassum, M. M. Mabrur, K. Yu, and
Commun. Technol., vol. 3, no. 3, pp. 1320–1331, Jul. M. Arifuzzaman, “Sentiment Analysis of Amazon
2023, doi: 10.48175/IJARSCT-14000T. Product Reviews Using Machine Learning and Deep
[23]. H. Sinha, “ANALYZING MOVIE REVIEW Learning Models,” in 2022 International Conference
SENTIMENTS ADVANCED MACHINE on Advancement in Electrical and Electronic
LEARNING AND NATURAL LANGUAGE Engineering, ICAEEE 2022, 2022. doi:
PROCESSING METHODS,” Int. Res. J. Mod. Eng. 10.1109/ICAEEE54957.2022.9836420.
Technol. Sci. (, vol. 06, no. 08, pp. 1326–1337, 2024, [34]. S. Al-Hagree and G. Al-Gaphari, “Arabic Sentiment
doi: https://www.doi.org/10.56726/IRJMETS60895. Analysis Based Machine Learning for Measuring User
[24]. R. Tandon, “An Analysis Of COVID-19 Tweets Satisfaction with Banking Services’ Mobile
Sentiments Based On Large Language Models Applications : Comparative Study,” in 2022 2nd
(Llms),” Int. J. Res. Anal. Rev., vol. 11, no. 3, 2024. International Conference on Emerging Smart
[25]. N. Yadav, O. Kudale, A. Rao, S. Gupta, and A. Technologies and Applications, eSmarTA 2022, 2022.
Shitole, “Twitter Sentiment Analysis Using doi: 10.1109/eSmarTA56775.2022.9935486.
Supervised Machine Learning,” in Lecture Notes on [35]. P. K. Vishwakarma and M. R. Suyambu, “An Analysis
Data Engineering and Communications Technologies, of Engineering , Procurement And Construction ( EPC
2021. doi: 10.1007/978-981-15-9509-7_51. ) -Contracts Based on Renewable Energy,” IJSART,
[26]. A. and P. Khare, “Cloud Security Challenges : vol. 10, no. 10, 2024.
Implementing Best Practices for Secure SaaS [36]. R. Goyal, “EXPLORING THE PERFORMANCE OF
Application Development,” Int. J. Curr. Eng. MACHINE LEARNING MODELS FOR
Technol., vol. 11, no. 6, pp. 669–676, 2021, doi: CLASSIFICATION AND IDENTIFICATION OF
https://doi.org/10.14741/ijcet/v.11.6.11. FRAUDULENT INSURANCE CLAIMS,” Int. J.
[27]. R. Tandon, “Social Media Texts Classification and Core Eng. Manag., vol. 7, no. 10, 2024.
Analysis Based on Large Language Models (LLMs),” [37]. P. Khare and S. Srivastava, “AI-Powered Fraud
J. Emerg. Technol. Innov. Res., vol. 11, no. 9, pp. Prevention: A Comprehensive Analysis of Machine
a751–a757, 2024. Learning Applications in Online Transactions,” J.
[28]. B. Ihnaini, W. Sun, Y. Cai, Z. Xu, and R. Sangi, Emerg. Technol. Innov. Res., vol. 10, pp. f518–f525,
“Sentiment Analysis of Song Dynasty Classical 2023.
Poetry Using Fine-Tuned Large Language Models: A [38]. R. K. Arora, Mohd.Muqeem, and M. Saxena,
Study with LLMs,” in 2024 7th International “Developing a Comprehensive Security Framework
Conference on Artificial Intelligence and Big Data for Detecting and Mitigating IoT device Attack,” Sep.
(ICAIBD), 2024, pp. 590–597. doi: 30, 2024. doi: 10.21203/rs.3.rs-5165811/v1.
10.1109/ICAIBD62003.2024.10604440. [39]. N. G. Abhinav Parashar, “Asset Master Data
[29]. H. A. Patrick, J. Praveen Gujjar, M. H. Sharief, and U. Management: Ensuring Accuracy and Consistency in
Mukherjee, “Sentiment Analysis Perspective using Industrial Operations,” IJNRD - Int. J. Nov. Res. Dev.,
Supervised Machine Learning Method,” in 2023 5th vol. 9, no. 9, pp. 861-a867, 2024.
International Conference on Electrical, Computer and [40]. S. Bauskar, “Enhancing System Observability with
Communication Technologies, ICECCT 2023, 2023. Machine Learning Techniques for Anomaly
doi: 10.1109/ICECCT56650.2023.10179807. Detection,” Int. J. Manag. IT Eng., vol. 14, no. 10, pp.
64–70, 2024.

IJISRT24DEC1694 www.ijisrt.com 1655

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886
[41]. R. Arora, S. Gera, and M. Saxena, “Mitigating [53]. Muthuvel Raj Suyambu and Pawan Kumar
security risks on privacy of sensitive data used in Vishwakarma, “Improving grid reliability with grid-
cloud-based ERP applications,” in Proceedings of the scale Battery Energy Storage Systems (BESS),” Int. J.
2021 8th International Conference on Computing for Sci. Res. Arch., vol. 13, no. 1, pp. 776–789, Sep. 2024,
Sustainable Global Development, INDIACom 2021, doi: 10.30574/ijsra.2024.13.1.1694.
2021. doi: 10.1109/INDIACom51348.2021.00081. [54]. P. Rajapaksha, R. Farahbakhsh, and N. Crespi,
[42]. S. Bauskar, “BUSINESS ANALYTICS IN “BERT, XLNet or RoBERTa: The Best Transfer
ENTERPRISE SYSTEM BASED ON Learning Model to Detect Clickbaits,” IEEE Access,
APPLICATION OF ARTIFICIAL 2021, doi: 10.1109/ACCESS.2021.3128742.
INTELLIGENCE,” Int. Res. J. Mod. Eng. Technol. [55]. R. Tandon, “Face mask detection model based on deep
Sci., vol. 04, no. 01, pp. 1861–1870, 2022, doi: DOI : CNN techniques using AWS,” Int. J. Eng. Res. Appl.,
https://www.doi.org/10.56726/IRJMETS18127. vol. 13, no. 5, pp. 12–19, 2023, doi: 10.9790/9622-
[43]. V. V. Kumar, A. Sahoo, and F. W. Liou, “Cyber- 13051219.
enabled Product Lifecycle Management: A Multi- [56]. N. Abid, “Enhanced IoT Network Security with
agent Framework,” Procedia Manuf., vol. 39, pp. 123– Machine Learning Techniques for Anomaly Detection
131, 2019, doi: 10.1016/j.promfg.2020.01.247. and Classification,” Int. J. Curr. Eng. Technol., vol.
[44]. Rajesh Goyal, “THE ROLE OF BUSINESS 13, no. 6, pp. 536–544, 2023, doi:
ANALYSTS IN INFORMATION MANAGEMENT https://doi.org/10.14741/ijcet/v.13.6.5.
PROJECTS,” Int. J. Core Eng. Manag., vol. 6, no. 9, [57]. S. Arora and A. Tewari, “AI-Driven Resilience :
2020. Enhancing Critical Infrastructure with Edge
[45]. N. G. Singh, Abhinav Parashar A, “Streamlining Computing,” Int. J. Curr. Eng. Technol., vol. 12, no.
Purchase Requisitions and Orders : A Guide to 2, pp. 151–157, 2022, doi:
Effective Goods Receipt Management,” J. Emerg. https://doi.org/10.14741/ijcet/v.12.2.9.
Technol. Innov. Res., vol. 8, no. 5, 2021. [58]. M. Z. Hasan, R. Fink, M. R. Suyambu, M. K.
[46]. V. K. Yarlagadda, “Cutting-edge developments in Baskaran, D. James, and J. Gamboa, “Performance
Robotics for Smart Warehousing and Logistics evaluation of energy efficient intelligent elevator
Optimization,” Robot. Xplore USA Tech Dig., vol. 1, controllers,” in IEEE International Conference on
no. 1, pp. 61–79, 2024. Electro Information Technology, 2015. doi:
[47]. A. P. A. Singh, “Best Practices for Creating and 10.1109/EIT.2015.7293320.
Maintaining Material Master Data in Industrial [59]. R. Goyal, “THE ROLE OF REQUIREMENT
Systems,” Int. J. Res. Anal. Rev., vol. 10, no. 1, pp. GATHERING IN AGILE SOFTWARE
112–119, 2023. DEVELOPMENT: STRATEGIES FOR SUCCESS
[48]. S. Pandya, “Predictive Analytics in Smart Grids : AND CHALLENGES,” Int. J. Core Eng. Manag., vol.
Leveraging Machine Learning for Renewable Energy 6, no. 12, pp. 142–152, 2021.
Sources,” Int. J. Curr. Eng. Technol., vol. 11, no. 6, [60]. Prity Choudhary, Rahul Choudhary and S. Garaga,
pp. 677–683, 2021, doi: “Enhancing Training by Incorporating ChatGPT in
https://doi.org/10.14741/ijcet/v.11.6.12. Learning Modules: An Exploration of Benefits,
[49]. P. K. and S. Arora, “the Impact of Machine Learning Challenges, and Best Practices,” Int. J. Innov. Sci.
and Ai on Enhancing Risk-Based Identity Verification Res. Technol., vol. 9, no. 11, 2024.
Processes,” Int. Res. J. Mod. Eng. Technol. Sci., vol. [61]. K. Ullah et al., “Short-Term Load Forecasting: A
06, no. 05, pp. 8246–8255, 2024, doi: Comprehensive Review and Simulation Study With
10.56726/IRJMETS57063. CNN-LSTM Hybrids Approach,” IEEE Access, vol.
[50]. B. Boddu, “Serverless Databases Are the Future of 12, pp. 111858–111881, 2024, doi:
Database Management,” 10.1109/ACCESS.2024.3440631.
https://jsaer.com/download/vol-6-iss-1- [62]. R. Tandon, “Prediction of Stock Market Trends Based
2019/JSAER2019-6-1-277-282.pdf, vol. 6, no. 1, p. 5, on Large Language Models,” J. Emerg. Technol.
2020. Innov. Res., vol. 11, no. 9, pp. a615–a622, 2024.
[51]. A. P. A. Singh, “STRATEGIC APPROACHES TO [63]. M. H. Hashmi, M. Affan, and R. Tandon, “A
MATERIALS DATA COLLECTION AND Customize Battery Management Approach for
INVENTORY MANAGEMENT,” Int. J. Bus. Quant. Satellite,” in Proceedings of the 2023 24th
Econ. Appl. Manag. Res., vol. 7, no. 5, 2022. International Carpathian Control Conference, ICCC
[52]. P. Khare, S. Arora, and S. Gupta, “Integration of 2023, 2023. doi:
Artificial Intelligence (AI) and Machine Learning 10.1109/ICCC57093.2023.10178893.
(ML) into Product Roadmap Planning,” in 2024 First [64]. Ramesh Bishukarma, “Privacy-preserving based
International Conference on Electronics, encryption techniques for securing data in cloud
Communication and Signal Processing (ICECSP), computing environments,” Int. J. Sci. Res. Arch., vol.
IEEE, Aug. 2024, pp. 1–6. doi: 9, no. 2, pp. 1014–1025, Aug. 2023, doi:
10.1109/ICECSP61809.2024.10698502. 10.30574/ijsra.2023.9.2.0441.

IJISRT24DEC1694 www.ijisrt.com 1656

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14575886
[65]. Sahil Arora and Apoorva Tewari, “Fortifying Critical
Infrastructures: Secure Data Management with Edge
Computing,” Int. J. Adv. Res. Sci. Commun. Technol.,
vol. 3, no. 2, pp. 946–955, Aug. 2023, doi:
10.48175/IJARSCT-12743E.
[66]. S. Bauskar, “ADVANCED ENCRYPTION
TECHNIQUES FOR ENHANCING DATA
SECURITY IN CLOUD COMPUTING
ENVIRONMENT,” Int. Res. J. Mod. Eng. Technol.
Sci., vol. 05, no. 10, 2023, doi:
10.56726/IRJMETS45283.
[67]. H. Sinha, “Benchmarking Predictive Performance Of
Machine Learning Approaches For Accurate
Prediction Of Boston House Prices : An In-Depth
Analysis,” iternational J. Res. Anal. Rev., vol. 11, no.
3, 2024.
[68]. M. R. S. and P. K. Vishwakarma, “THE
ASSESSMENTS OF FINANCIAL RISK BASED
ON RENEWABLE ENERGY INDUSTRY,” Int. Res.
J. Mod. Eng. Technol. Sci., vol. 06, no. 09, pp. 758–
770, 2024.
[69]. H. Sinha, “Predicting Employee Performance in
Business Environments Using Effective Machine
Learning Models,” IJNRD - Int. J. Nov. Res. Dev.,
vol. 9, no. 9, pp. a875–a881, 2024.
[70]. R. Bishukarma, “Optimising Cloud Security in Multi-
Cloud Environments : A Study of Best Practices,” Int.
Res. J., vol. 11, no. 11, pp. 590–598, 2024.
[71]. H. Cam, A. V. Cam, U. Demirel, and S. Ahmed,
“Sentiment analysis of financial Twitter posts on
Twitter with the machine learning classifiers,”
Heliyon, 2024, doi: 10.1016/j.heliyon.2023.e23784.
[72]. T. Zahra and I. Hussain, “Sentiment Analysis of
Twitter Dataset Using Lle and Classification
Methods,” no. 01, pp. 1151–1164, 2021.
[73]. K. Deepa, H. Sangita, and H. Shruthi, “Sentiment
Analysis of Twitter Data Using Machine Learning,”
Lect. Notes Electr. Eng., vol. 905, no. June, pp. 259–
267, 2022, doi: 10.1007/978-981-19-2177-3_26.

IJISRT24DEC1694 www.ijisrt.com 1657